diff --git a/.circleci/config.yml b/.circleci/config.yml
index 55fa9410845..dfc539fb80e 100644
--- a/.circleci/config.yml
+++ b/.circleci/config.yml
@@ -112,10 +112,10 @@ commands:
             node --version
             npm --version
   install_rust:
-    description: "Install pinned rustup (1.28.2) and Rust toolchain (1.97.1) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
+    description: "Install pinned rustup (1.28.2) and Rust toolchain (1.98.0) with checksum verification. Adds ~/.cargo/bin to PATH. Run this before any `uv sync` or `uv build` of the workspace: the root package builds litellm-rust through maturin, and on an image without cargo maturin fetches an unpinned rustup and a floating toolchain by itself."
     steps:
       - run:
-          name: Install Rust (rustup 1.28.2, toolchain 1.97.1)
+          name: Install Rust (rustup 1.28.2, toolchain 1.98.0)
           command: |
             case "$(uname -m)" in
               x86_64)
@@ -135,7 +135,7 @@ commands:
               "https://static.rust-lang.org/rustup/archive/1.28.2/${RUSTUP_TRIPLE}/rustup-init"
             echo "${RUSTUP_SHA256}  /tmp/rustup-init" | sha256sum -c -
             chmod +x /tmp/rustup-init
-            /tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.97.1
+            /tmp/rustup-init -y --no-modify-path --profile minimal --default-toolchain 1.98.0
             rm -f /tmp/rustup-init
             echo 'export PATH="$HOME/.cargo/bin:$PATH"' >> "$BASH_ENV"
             export PATH="$HOME/.cargo/bin:$PATH"
@@ -300,7 +300,7 @@ jobs:
             if ($rustupActual -ne $rustupExpected) {
               throw "rustup installer hash mismatch: expected $rustupExpected got $rustupActual"
             }
-            & $rustupInit -y --profile minimal --default-toolchain stable
+            & $rustupInit -y --profile minimal --default-toolchain 1.98.0
             if ($LASTEXITCODE -ne 0) {
               exit $LASTEXITCODE
             }
diff --git a/.github/scripts/smoke_test_native_wheel.py b/.github/scripts/smoke_test_native_wheel.py
new file mode 100644
index 00000000000..577bb32fcf0
--- /dev/null
+++ b/.github/scripts/smoke_test_native_wheel.py
@@ -0,0 +1,68 @@
+from __future__ import annotations
+
+import subprocess
+import sys
+import tempfile
+import zipfile
+from pathlib import Path
+from typing import Final
+
+CHILD_SCRIPT: Final = """
+from importlib.util import module_from_spec, spec_from_file_location
+from pathlib import Path
+import sys
+
+native_path = Path(sys.argv[1])
+spec = spec_from_file_location("litellm.rust_bridge._native", native_path)
+if spec is None or spec.loader is None:
+    raise RuntimeError("cannot create native extension import specification")
+module = module_from_spec(spec)
+spec.loader.exec_module(module)
+
+before = module.gil_stats()
+if not isinstance(before.get("releases"), int):
+    raise AssertionError(f"unexpected gil_stats result: {before!r}")
+
+try:
+    module._panic_for_test()
+except BaseException as error:
+    if type(error).__name__ != "PanicException":
+        raise AssertionError(f"expected PanicException, got {type(error).__name__}") from error
+else:
+    raise AssertionError("Rust panic returned without raising")
+
+after = module.gil_stats()
+if not isinstance(after.get("releases"), int):
+    raise AssertionError(f"native module unusable after panic: {after!r}")
+"""
+
+
+def main() -> int:
+    if len(sys.argv) != 2:
+        sys.stderr.write(f"usage: {Path(sys.argv[0]).name} WHEEL\n")
+        return 2
+
+    wheel: Final = Path(sys.argv[1])
+    with tempfile.TemporaryDirectory() as temporary_directory, zipfile.ZipFile(wheel) as archive:
+        native_members: Final = tuple(
+            member
+            for member in archive.infolist()
+            if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
+        )
+        if len(native_members) != 1:
+            sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
+            return 1
+
+        native_path: Final = Path(temporary_directory) / Path(native_members[0].filename).name
+        native_path.write_bytes(archive.read(native_members[0]))
+        result: Final = subprocess.run((sys.executable, "-c", CHILD_SCRIPT, str(native_path)), check=False)
+
+    if result.returncode != 0:
+        sys.stderr.write(f"native wheel smoke test exited with status {result.returncode}\n")
+        return 1
+
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.github/scripts/verify_linux_native_wheel.py b/.github/scripts/verify_linux_native_wheel.py
new file mode 100644
index 00000000000..899e2a211c0
--- /dev/null
+++ b/.github/scripts/verify_linux_native_wheel.py
@@ -0,0 +1,282 @@
+from __future__ import annotations
+
+import importlib.util
+import os
+import re
+import subprocess
+import sys
+import zipfile
+from collections.abc import Callable, Mapping, Sequence
+from itertools import product
+from pathlib import Path, PurePosixPath
+from types import MappingProxyType, ModuleType
+from typing import Final, Protocol
+
+EXPECTED_PYTHON_TAG: Final = "cp310"
+EXPECTED_ABI_TAG: Final = "abi3"
+EXPECTED_PLATFORM_TAG: Final = "linux_x86_64"
+
+
+class CommandRunner(Protocol):
+    def __call__(
+        self,
+        command: tuple[str, ...],
+        *,
+        check: bool,
+        capture_output: bool,
+        text: bool,
+    ) -> subprocess.CompletedProcess[str]: ...
+
+
+def _run_command(
+    command: tuple[str, ...],
+    *,
+    check: bool,
+    capture_output: bool,
+    text: bool,
+) -> subprocess.CompletedProcess[str]:
+    return subprocess.run(command, check=check, capture_output=capture_output, text=text)
+
+
+def _dist_info_directory(member: zipfile.ZipInfo) -> str | None:
+    parts: Final = PurePosixPath(member.filename).parts
+    if not parts or not parts[0].endswith(".dist-info"):
+        return None
+    return parts[0]
+
+
+def _wheel_metadata_tags(archive: zipfile.ZipFile, members: tuple[zipfile.ZipInfo, ...]) -> tuple[str, ...]:
+    if len(members) != 1:
+        return ()
+    lines: Final = archive.read(members[0]).splitlines()
+    return tuple(line.removeprefix(b"Tag:").strip().decode("ascii") for line in lines if line.startswith(b"Tag:"))
+
+
+def _load_native_module(native_path: Path) -> ModuleType | None:
+    module_spec: Final = importlib.util.spec_from_file_location("litellm.rust_bridge._native", native_path)
+    if module_spec is None or module_spec.loader is None:
+        return None
+    try:
+        native_module: Final = importlib.util.module_from_spec(module_spec)
+        module_spec.loader.exec_module(native_module)
+    except Exception as error:  # noqa: BLE001  # native module initialization can raise arbitrary exceptions
+        sys.stderr.write(f"native module load failed: {error}\n")
+        return None
+    return native_module
+
+
+def main(
+    argv: Sequence[str] | None = None,
+    environment: Mapping[str, str] | None = None,
+    load_native_module: Callable[[Path], ModuleType | None] = _load_native_module,
+    run_command: CommandRunner = _run_command,
+) -> int:
+    arguments: Final = tuple(sys.argv if argv is None else argv)
+    resolved_environment: Final = os.environ if environment is None else environment
+    if len(arguments) != 2:
+        sys.stderr.write(f"usage: {Path(arguments[0]).name} WHEEL\n")
+        return 2
+
+    wheel: Final = Path(arguments[1])
+    wheel_tags: Final = wheel.stem.rsplit("-", maxsplit=3)
+    if len(wheel_tags) != 4:
+        sys.stderr.write(f"cannot parse wheel tags from {wheel.name}\n")
+        return 1
+
+    wheel_identity: Final = wheel_tags[0].split("-")
+    if len(wheel_identity) != 2 or wheel_identity[0] != "litellm" or not wheel_identity[1]:
+        sys.stderr.write(f"unexpected wheel identity: {wheel_tags[0]}\n")
+        return 1
+
+    expected_dist_info_directory: Final = f"{wheel_tags[0]}.dist-info"
+    expected_dist_info_directories: Final = frozenset((expected_dist_info_directory,))
+    python_tag: Final = wheel_tags[1]
+    abi_tag: Final = wheel_tags[2]
+    platform_tag: Final = wheel_tags[3]
+    expanded_filename_tags: Final = frozenset(
+        "-".join(tag) for tag in product(python_tag.split("."), abi_tag.split("."), platform_tag.split("."))
+    )
+
+    with zipfile.ZipFile(wheel) as archive:
+        wheel_members: Final = archive.infolist()
+        dist_info_directories: Final = frozenset(
+            directory for member in wheel_members if (directory := _dist_info_directory(member)) is not None
+        )
+        required_dist_info_files: Final = ("METADATA", "RECORD", "WHEEL")
+        dist_info_file_counts: Final = MappingProxyType(
+            {
+                filename: sum(
+                    member.filename == f"{expected_dist_info_directory}/{filename}" for member in wheel_members
+                )
+                for filename in required_dist_info_files
+            }
+        )
+        wheel_metadata_members: Final = tuple(
+            member for member in wheel_members if member.filename == f"{expected_dist_info_directory}/WHEEL"
+        )
+        wheel_metadata_tags: Final = _wheel_metadata_tags(archive, wheel_metadata_members)
+        native_members: Final = tuple(
+            member
+            for member in wheel_members
+            if member.filename.startswith("litellm/rust_bridge/_native.") and member.filename.endswith(".so")
+        )
+        if len(native_members) != 1:
+            sys.stderr.write(f"expected one native extension, found {len(native_members)}\n")
+            return 1
+
+        unexpected_members: Final = tuple(
+            member.filename
+            for member in wheel_members
+            if member.filename.endswith((".pdb", ".dwp", ".rlib", ".rmeta", "Cargo.toml", "Cargo.lock"))
+            or any(part.endswith(".dSYM") for part in PurePosixPath(member.filename).parts)
+        )
+        native_member: Final = native_members[0]
+        uncompressed_wheel_size: Final = sum(member.file_size for member in wheel_members)
+        native_path: Final = wheel.parent / "native" / Path(native_member.filename).name
+        native_path.parent.mkdir(parents=True, exist_ok=True)
+        native_path.write_bytes(archive.read(native_member))
+
+    wheel_metadata_tags_match: Final = (
+        len(wheel_metadata_tags) == len(expanded_filename_tags)
+        and frozenset(wheel_metadata_tags) == expanded_filename_tags
+    )
+    commit_sha: Final = resolved_environment.get(
+        "RELEASE_WHEEL_COMMIT_SHA", resolved_environment.get("GITHUB_SHA", "unknown")
+    )
+    rustc_version: Final = run_command(
+        ("rustc", "--version"),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout.strip()
+    pyproject: Final = (Path(__file__).parents[2] / "pyproject.toml").read_text()
+    maturin_match: Final = re.search(r'"maturin==([^";]+)', pyproject)
+    if maturin_match is None:
+        sys.stderr.write("build-system does not pin an exact Maturin version\n")
+        return 1
+
+    maturin_version: Final = maturin_match.group(1)
+    native_percentage: Final = native_member.file_size / uncompressed_wheel_size * 100
+    size_report: Final = "\n".join(
+        (
+            "## Native wheel build report",
+            "",
+            "| Build | Value |",
+            "| --- | --- |",
+            f"| Commit | `{commit_sha}` |",
+            f"| Platform | `{platform_tag}` |",
+            f"| Python ABI | `{python_tag}-{abi_tag}` |",
+            f"| Rust compiler | `{rustc_version}` |",
+            f"| Maturin | `{maturin_version}` |",
+            "| Cargo profile | `release` |",
+            "",
+            "| Artifact | Size |",
+            "| --- | ---: |",
+            f"| Compressed wheel | {wheel.stat().st_size / 1_000_000:.2f} MB |",
+            f"| Uncompressed wheel | {uncompressed_wheel_size / 1_000_000:.2f} MB |",
+            f"| Native extension | {native_member.file_size / 1_000_000:.2f} MB |",
+            f"| Native share | {native_percentage:.2f}% |",
+            "",
+        )
+    )
+    summary_path: Final = resolved_environment.get("GITHUB_STEP_SUMMARY")
+    if summary_path is None:
+        sys.stdout.write(size_report)
+    else:
+        Path(summary_path).write_text(size_report)
+
+    sections: Final = run_command(
+        ("readelf", "--sections", "--wide", str(native_path)),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout
+    debug_sections: Final = tuple(section for section in (".debug_", ".zdebug_") if section in sections)
+    debug_sections_absent: Final = not debug_sections
+    static_symbol_table_absent: Final = ".symtab" not in sections
+
+    dynamic_symbols: Final = run_command(
+        ("readelf", "--dyn-syms", "--wide", str(native_path)),
+        check=True,
+        capture_output=True,
+        text=True,
+    ).stdout
+    extension_entry_point_present: Final = "PyInit__native" in dynamic_symbols
+    native_module: Final = load_native_module(native_path)
+    native_module_loads: Final = native_module is not None
+    panic_test_hook_absent: Final = native_module is not None and not hasattr(native_module, "_panic_for_test")
+    native_size_limit: Final = 20_000_000
+    native_size_within_limit: Final = native_member.file_size <= native_size_limit
+    validations: Final = (
+        (f"Python tag is {EXPECTED_PYTHON_TAG}", python_tag == EXPECTED_PYTHON_TAG),
+        (f"ABI tag is {EXPECTED_ABI_TAG}", abi_tag == EXPECTED_ABI_TAG),
+        (f"Platform tag is {EXPECTED_PLATFORM_TAG}", platform_tag == EXPECTED_PLATFORM_TAG),
+        ("Wheel dist-info directory matches the filename", dist_info_directories == expected_dist_info_directories),
+        (
+            "Required dist-info files are present exactly once",
+            all(count == 1 for count in dist_info_file_counts.values()),
+        ),
+        ("Wheel metadata tags match the filename", wheel_metadata_tags_match),
+        ("Debug sections are absent", debug_sections_absent),
+        ("Static symbol table is absent", static_symbol_table_absent),
+        ("Python extension entry point is present", extension_entry_point_present),
+        ("Native module loads", native_module_loads),
+        ("Production module omits the panic test hook", panic_test_hook_absent),
+        ("Native extension does not exceed 20 MB", native_size_within_limit),
+        ("Wheel contents are valid", not unexpected_members),
+    )
+
+    verified_report: Final = size_report + "\n".join(
+        ("", "| Validation | Expected | Result |", "| --- | --- | :---: |")
+        + tuple(f"| {label} | Yes | {'O' if passed else 'X'} |" for label, passed in validations)
+        + ("",)
+    )
+    if summary_path is not None:
+        Path(summary_path).write_text(verified_report)
+
+    invalid_dist_info_files: Final = any(count != 1 for count in dist_info_file_counts.values())
+    validation_errors: Final = tuple(
+        message
+        for failed, message in (
+            (bool(debug_sections), f"{native_member.filename} contains debug sections: {', '.join(debug_sections)}"),
+            (not static_symbol_table_absent, f"{native_member.filename} contains a static symbol table"),
+            (not extension_entry_point_present, "native extension does not export PyInit__native"),
+            (
+                python_tag != EXPECTED_PYTHON_TAG,
+                f"unexpected Python tag: expected {EXPECTED_PYTHON_TAG}, found {python_tag}",
+            ),
+            (abi_tag != EXPECTED_ABI_TAG, f"unexpected ABI tag: expected {EXPECTED_ABI_TAG}, found {abi_tag}"),
+            (
+                platform_tag != EXPECTED_PLATFORM_TAG,
+                f"unexpected platform tag: expected {EXPECTED_PLATFORM_TAG}, found {platform_tag}",
+            ),
+            (
+                dist_info_directories != expected_dist_info_directories,
+                f"unexpected dist-info directories: expected {expected_dist_info_directory}, "
+                f"found {', '.join(sorted(dist_info_directories))}",
+            ),
+            (invalid_dist_info_files, f"required dist-info file counts are invalid: {dist_info_file_counts}"),
+            (
+                not invalid_dist_info_files and not wheel_metadata_tags_match,
+                f"WHEEL tags do not match filename: expected {', '.join(sorted(expanded_filename_tags))}, "
+                f"found {', '.join(sorted(wheel_metadata_tags))}",
+            ),
+            (
+                native_module is not None and not panic_test_hook_absent,
+                "production native module exposes _panic_for_test",
+            ),
+            (
+                not native_size_within_limit,
+                f"native extension exceeds 20 MB: {native_member.file_size / 1_000_000:.2f} MB",
+            ),
+            (bool(unexpected_members), f"wheel contains unexpected build artifacts: {', '.join(unexpected_members)}"),
+        )
+        if failed
+    )
+    sys.stderr.write("".join(f"{message}\n" for message in validation_errors))
+
+    return 0 if all(passed for _, passed in validations) else 1
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/.github/workflows/report-rust-release-wheel.yml b/.github/workflows/report-rust-release-wheel.yml
new file mode 100644
index 00000000000..1d93b56f77f
--- /dev/null
+++ b/.github/workflows/report-rust-release-wheel.yml
@@ -0,0 +1,130 @@
+name: Report LiteLLM Rust release wheel
+
+on: # zizmor: ignore[dangerous-triggers] reporter executes no PR code and consumes no PR artifacts or outputs
+  workflow_run:
+    workflows:
+      - LiteLLM Rust
+    types:
+      - completed
+
+permissions: {}
+
+concurrency:
+  group: ${{ github.workflow }}-${{ github.event.workflow_run.pull_requests[0].number || github.event.workflow_run.id }}
+  cancel-in-progress: false
+
+jobs:
+  report-release-wheel:
+    name: report release wheel
+    if: >-
+      github.event.workflow_run.event == 'pull_request' &&
+      github.event.workflow_run.path == '.github/workflows/test-rust.yml' &&
+      github.event.workflow_run.head_repository.full_name == github.repository &&
+      github.event.workflow_run.pull_requests[0].number != null
+    runs-on: ubuntu-latest
+    timeout-minutes: 5
+    permissions:
+      issues: write # PR comments use the issues API
+      pull-requests: read # Current-head validation rejects stale workflow runs
+
+    steps:
+      - name: Link release wheel report on PR
+        uses: actions/github-script@f28e40c7f34bde8b3046d885e986cb6290c5673b # v7.1.0
+        env:
+          COMMENT_MARKER: "<!-- litellm-release-wheel-size -->"
+        with:
+          script: |
+            const marker = process.env.COMMENT_MARKER;
+            const workflowRun = context.payload.workflow_run;
+            const allowedConclusions = new Set([
+              "action_required",
+              "cancelled",
+              "failure",
+              "neutral",
+              "skipped",
+              "stale",
+              "startup_failure",
+              "success",
+              "timed_out",
+            ]);
+            if (
+              !allowedConclusions.has(workflowRun.conclusion) ||
+              workflowRun.event !== "pull_request" ||
+              workflowRun.path !== ".github/workflows/test-rust.yml" ||
+              workflowRun.head_repository?.full_name !==
+                `${context.repo.owner}/${context.repo.repo}` ||
+              workflowRun.pull_requests?.length !== 1
+            ) {
+              throw new Error("unexpected source workflow");
+            }
+            const pullRequest = workflowRun.pull_requests[0];
+            const pullRequestNumber = pullRequest.number;
+            const headSha = workflowRun.head_sha;
+            const runId = workflowRun.id;
+            if (
+              !Number.isSafeInteger(pullRequestNumber) ||
+              pullRequestNumber <= 0 ||
+              !Number.isSafeInteger(runId) ||
+              runId <= 0 ||
+              !/^[0-9a-f]{40}$/.test(headSha) ||
+              pullRequest.head?.sha !== headSha
+            ) {
+              throw new Error("invalid source workflow metadata");
+            }
+            const runUrl =
+              `${context.serverUrl}/${context.repo.owner}/${context.repo.repo}` +
+              `/actions/runs/${runId}`;
+            const result =
+              workflowRun.conclusion === "success"
+                ? "successfully"
+                : `with \`${workflowRun.conclusion}\``;
+            const body = [
+              marker,
+              "## LiteLLM Rust workflow",
+              "",
+              `Workflow completed ${result} for \`${headSha}\``,
+              "",
+              `[View workflow run](${runUrl})`,
+            ].join("\n");
+            const comments = await github.paginate(github.rest.issues.listComments, {
+              owner: context.repo.owner,
+              repo: context.repo.repo,
+              issue_number: pullRequestNumber,
+              per_page: 100,
+            });
+            const existing = comments.find(
+              (comment) =>
+                comment.user?.login === "github-actions[bot]" &&
+                comment.body?.startsWith(marker),
+            );
+            const currentPullRequest = (
+              await github.rest.pulls.get({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                pull_number: pullRequestNumber,
+              })
+            ).data;
+            if (
+              currentPullRequest.state !== "open" ||
+              currentPullRequest.head.repo?.full_name !==
+                `${context.repo.owner}/${context.repo.repo}` ||
+              currentPullRequest.head.sha !== headSha
+            ) {
+              core.info("source workflow no longer matches the current pull request head");
+              return;
+            }
+            if (existing) {
+              await github.rest.issues.updateComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                comment_id: existing.id,
+                body,
+              });
+            } else {
+              await github.rest.issues.createComment({
+                owner: context.repo.owner,
+                repo: context.repo.repo,
+                issue_number: pullRequestNumber,
+                body,
+              });
+            }
diff --git a/.github/workflows/test-rust.yml b/.github/workflows/test-rust.yml
index 21e1bcb90c6..aada0fcf239 100644
--- a/.github/workflows/test-rust.yml
+++ b/.github/workflows/test-rust.yml
@@ -4,6 +4,11 @@ on:
   push:
     paths:
       - "litellm-rust/**"
+      - ".cargo/**"
+      - "pyproject.toml"
+      - "rust-toolchain.toml"
+      - ".github/scripts/smoke_test_native_wheel.py"
+      - ".github/scripts/verify_linux_native_wheel.py"
       - ".github/workflows/test-rust.yml"
   pull_request:
     branches:
@@ -13,6 +18,11 @@ on:
       - "litellm_**"
     paths:
       - "litellm-rust/**"
+      - ".cargo/**"
+      - "pyproject.toml"
+      - "rust-toolchain.toml"
+      - ".github/scripts/smoke_test_native_wheel.py"
+      - ".github/scripts/verify_linux_native_wheel.py"
       - ".github/workflows/test-rust.yml"
 
 permissions:
@@ -40,9 +50,7 @@ jobs:
           persist-credentials: false
 
       - name: Set up Rust
-        run: |
-          rustup toolchain install stable --profile minimal --component clippy,rustfmt
-          rustup default stable
+        run: rustup toolchain install
 
       - name: Cache Cargo registry and target
         uses: actions/cache@0057852bfaa89a56745cba8c7296529d2fc39830 # v4.3.0
@@ -51,7 +59,7 @@ jobs:
             ~/.cargo/registry
             ~/.cargo/git
             litellm-rust/target
-          key: ${{ runner.os }}-cargo-${{ hashFiles('litellm-rust/Cargo.lock') }}
+          key: ${{ runner.os }}-cargo-${{ hashFiles('rust-toolchain.toml', 'litellm-rust/Cargo.lock') }}
           restore-keys: |
             ${{ runner.os }}-cargo-
 
@@ -69,3 +77,47 @@ jobs:
 
       - name: Run core tests with Bedrock auth
         run: cargo test -p litellm-core --features bedrock-auth --locked
+
+  release-wheel:
+    name: release wheel
+    runs-on: ubuntu-latest
+    timeout-minutes: 20
+    permissions:
+      contents: read
+    env:
+      CARGO_TERM_COLOR: always
+
+    steps:
+      - name: Checkout repository
+        uses: actions/checkout@08eba0b27e820071cde6df949e0beb9ba4906955 # v4.3.0
+        with:
+          persist-credentials: false
+
+      - name: Set up Python
+        uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5.6.0
+        with:
+          python-version: "3.12"
+
+      - name: Set up uv
+        uses: ./.github/actions/setup-uv-with-retries
+        with:
+          version: "0.10.9"
+
+      - name: Set up Rust
+        run: rustup toolchain install
+
+      - name: Build release wheel
+        run: uv build --wheel --out-dir dist
+
+      - name: Build panic contract wheel
+        run: >-
+          uv build --wheel --out-dir panic-dist
+          --config-setting "maturin.build-args=--features panic-test,extension-module"
+
+      - name: Smoke-test native panic unwinding
+        run: python .github/scripts/smoke_test_native_wheel.py panic-dist/*.whl
+
+      - name: Verify stripped native extension
+        env:
+          RELEASE_WHEEL_COMMIT_SHA: ${{ github.event.pull_request.head.sha || github.sha }}
+        run: python .github/scripts/verify_linux_native_wheel.py dist/*.whl
diff --git a/backend/Dockerfile b/backend/Dockerfile
index aa01b9fba8b..622fedcd70d 100644
--- a/backend/Dockerfile
+++ b/backend/Dockerfile
@@ -46,6 +46,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
         --extra proxy-runtime \
         --extra extra_proxy \
         --extra semantic-router \
+        --extra saml \
         --python python3.13
 
 # Stage 2 — copy source and install the project + workspace members.
@@ -57,6 +58,7 @@ RUN --mount=type=cache,target=/root/.cache/uv \
         --extra proxy-runtime \
         --extra extra_proxy \
         --extra semantic-router \
+        --extra saml \
         --python python3.13
 
 RUN HOME=/opt/prisma XDG_CACHE_HOME=/opt/prisma/.cache PRISMA_BINARY_CACHE_DIR=/opt/prisma/binaries \
diff --git a/helm/litellm-helm/Chart.yaml b/helm/litellm-helm/Chart.yaml
index 3959d85edf3..a3cb388ffc6 100644
--- a/helm/litellm-helm/Chart.yaml
+++ b/helm/litellm-helm/Chart.yaml
@@ -18,7 +18,7 @@ type: application
 # This is the chart version. This version number should be incremented each time you make changes
 # to the chart and its templates, including the app version.
 # Versions are expected to follow Semantic Versioning (https://semver.org/)
-version: 1.1.2
+version: 1.1.3
 
 # This is the version number of the application being deployed. This version number should be
 # incremented each time you make changes to the application. Versions are not expected to
diff --git a/helm/litellm-helm/tests/hpa_tests.yaml b/helm/litellm-helm/tests/hpa_tests.yaml
index ec18c3591d3..cd062dd5971 100644
--- a/helm/litellm-helm/tests/hpa_tests.yaml
+++ b/helm/litellm-helm/tests/hpa_tests.yaml
@@ -1,4 +1,4 @@
-suite: "hpa with behavior"
+suite: "hpa"
 templates:
   - hpa.yaml
 tests:
@@ -23,14 +23,44 @@ tests:
       - equal: { path: spec.behavior.scaleUp.stabilizationWindowSeconds, value: 60 }
       - equal: { path: spec.behavior.scaleDown.stabilizationWindowSeconds, value: 90 }
 
----
-suite: "hpa without behavior"
-templates:
-  - hpa.yaml
-tests:
   - it: "does not render behavior when not set"
     set:
       autoscaling.enabled: true
     asserts:
       - isKind: { of: HorizontalPodAutoscaler }
       - isNull: { path: spec.behavior }
+
+  - it: "scales on cpu at the documented 60 percent by default"
+    set:
+      autoscaling.enabled: true
+    asserts:
+      - isKind: { of: HorizontalPodAutoscaler }
+      - equal: { path: "spec.metrics[0].resource.name", value: cpu }
+      - equal: { path: "spec.metrics[0].resource.target.type", value: Utilization }
+      - equal: { path: "spec.metrics[0].resource.target.averageUtilization", value: 60 }
+
+  - it: "does not scale on memory by default"
+    set:
+      autoscaling.enabled: true
+    asserts:
+      - lengthEqual: { path: spec.metrics, count: 1 }
+
+  - it: "honours an explicit cpu target override"
+    set:
+      autoscaling.enabled: true
+      autoscaling.targetCPUUtilizationPercentage: 75
+    asserts:
+      - equal: { path: "spec.metrics[0].resource.target.averageUtilization", value: 75 }
+
+  - it: "renders a memory metric only when a memory target is set"
+    set:
+      autoscaling.enabled: true
+      autoscaling.targetMemoryUtilizationPercentage: 80
+    asserts:
+      - lengthEqual: { path: spec.metrics, count: 2 }
+      - equal: { path: "spec.metrics[1].resource.name", value: memory }
+      - equal: { path: "spec.metrics[1].resource.target.averageUtilization", value: 80 }
+
+  - it: "renders no hpa when autoscaling is disabled"
+    asserts:
+      - hasDocuments: { count: 0 }
diff --git a/helm/litellm-helm/values.yaml b/helm/litellm-helm/values.yaml
index f8df98de102..637be2322e3 100644
--- a/helm/litellm-helm/values.yaml
+++ b/helm/litellm-helm/values.yaml
@@ -200,7 +200,16 @@ autoscaling:
   enabled: false
   minReplicas: 1
   maxReplicas: 100
-  targetCPUUtilizationPercentage: 80
+  # 60 is the documented recommendation. See "Recommended Machine Specifications"
+  # in https://docs.litellm.ai/docs/proxy/prod. A new replica clears the startupProbe
+  # above only after up to failureThreshold x periodSeconds = 300 seconds, so a target
+  # high enough to trip near saturation adds capacity minutes after it was needed.
+  targetCPUUtilizationPercentage: 60
+  # Deliberately left unset rather than given a value. The prisma query engine's
+  # resident memory is a high-water mark that ratchets to the pod's worst-ever write
+  # and is never returned, so a memory target reads the largest write a pod ever did
+  # rather than what it is doing now, and replicas ratchet up without scaling back in.
+  # Memory is a floor to provision under 'resources', not a signal to scale on.
   # targetMemoryUtilizationPercentage: 80
   # behavior: {}
 
diff --git a/litellm-proxy-extras/litellm_proxy_extras/prisma_toolchain.py b/litellm-proxy-extras/litellm_proxy_extras/prisma_toolchain.py
index f3b55fd4d96..2283814ab35 100644
--- a/litellm-proxy-extras/litellm_proxy_extras/prisma_toolchain.py
+++ b/litellm-proxy-extras/litellm_proxy_extras/prisma_toolchain.py
@@ -14,9 +14,22 @@ then fails on a Node binary that was never written. Deleting a cache directory
 that exists without a Node binary is what turns a killed bootstrap back into a
 recoverable one.
 
-Both budgets are overridable so an operator can widen them without a release:
-``LITELLM_PRISMA_BOOTSTRAP_TIMEOUT`` for the toolchain install and
-``LITELLM_PRISMA_COMMAND_TIMEOUT`` for every individual Prisma command.
+``prisma migrate deploy`` is the other command whose runtime is not a
+constant: it grows with the number of pending migrations, so a fresh database
+that has to replay every migration this package ships overruns a per-command
+budget sized for the short bookkeeping commands, on a laptop as much as on a
+slow CI runner. The Python ``prisma`` wrapper spawns Node and the schema engine
+as separate children, so killing the wrapper on timeout leaves them running:
+the retry then contends with that orphan for Prisma's advisory lock and cannot
+finish any sooner. Migrate deploy therefore runs under its own budget.
+
+All three budgets are overridable so an operator can widen them without a
+release: ``LITELLM_PRISMA_BOOTSTRAP_TIMEOUT`` for the toolchain install,
+``LITELLM_PRISMA_MIGRATE_DEPLOY_TIMEOUT`` for ``prisma migrate deploy`` and
+``LITELLM_PRISMA_COMMAND_TIMEOUT`` for every other Prisma command. The
+per-command budget used to bound migrate deploy as well, so a deployment that
+raised it above the deploy default keeps that larger budget for deploy unless
+the deploy override says otherwise.
 """
 
 import math
@@ -36,10 +49,12 @@ except ImportError:
 
 PRISMA_COMMAND_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_COMMAND_TIMEOUT"
 PRISMA_BOOTSTRAP_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_BOOTSTRAP_TIMEOUT"
+PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR = "LITELLM_PRISMA_MIGRATE_DEPLOY_TIMEOUT"
 NODEENV_CACHE_DIR_ENV_VAR = "PRISMA_NODEENV_CACHE_DIR"
 
 DEFAULT_PRISMA_COMMAND_TIMEOUT = 60.0
 DEFAULT_PRISMA_BOOTSTRAP_TIMEOUT = 600.0
+DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT = 600.0
 
 BOOTSTRAP_ARG = "--version"
 
@@ -88,6 +103,15 @@ def prisma_bootstrap_timeout() -> float:
     )
 
 
+def prisma_migrate_deploy_timeout() -> float:
+    """Seconds one ``prisma migrate deploy`` may run for, however many migrations are pending."""
+    if os.getenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR) is not None:
+        return _timeout_from_env(
+            PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT
+        )
+    return max(DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT, prisma_command_timeout())
+
+
 def nodeenv_cache_dir() -> Optional[Path]:
     """Where Prisma installs its private Node runtime, or None if unknowable."""
     override = os.getenv(NODEENV_CACHE_DIR_ENV_VAR)
diff --git a/litellm-proxy-extras/litellm_proxy_extras/utils.py b/litellm-proxy-extras/litellm_proxy_extras/utils.py
index b8032dd0d28..f6647268624 100644
--- a/litellm-proxy-extras/litellm_proxy_extras/utils.py
+++ b/litellm-proxy-extras/litellm_proxy_extras/utils.py
@@ -15,8 +15,11 @@ from litellm_proxy_extras.replica_identity import (
     apply_replica_identity_full,
 )
 from litellm_proxy_extras.prisma_toolchain import (
+    PRISMA_COMMAND_TIMEOUT_ENV_VAR,
+    PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR,
     ensure_prisma_toolchain,
     prisma_command_timeout,
+    prisma_migrate_deploy_timeout,
 )
 
 
@@ -698,12 +701,13 @@ class ProxyExtrasDBManager:
 
         original_dir = os.getcwd()
         os.chdir(migrations_dir)
+        deploy_timeout = prisma_migrate_deploy_timeout()
         try:
             for attempt in range(4):
                 try:
                     result = subprocess.run(
                         [_get_prisma_command(), "migrate", "deploy"],
-                        timeout=prisma_command_timeout(),
+                        timeout=deploy_timeout,
                         check=True,
                         capture_output=True,
                         text=True,
@@ -713,8 +717,12 @@ class ProxyExtrasDBManager:
                     return True
 
                 except subprocess.TimeoutExpired:
-                    logger.info(
-                        f"prisma migrate deploy attempt {attempt + 1} timed out, retrying"
+                    logger.warning(
+                        "prisma migrate deploy attempt %s timed out after %ss, retrying. "
+                        "Raise %s if this database needs longer to apply its pending migrations.",
+                        attempt + 1,
+                        deploy_timeout,
+                        PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR,
                     )
                     time.sleep(random.randrange(5, 15))
                     continue
@@ -823,7 +831,8 @@ class ProxyExtrasDBManager:
                 "Database migration failed after 4 attempts (retry loop "
                 "exhausted by timeouts or repeated idempotent-recovery "
                 "continues). Check database connectivity, load, and "
-                "_prisma_migrations ledger state."
+                "_prisma_migrations ledger state, and raise "
+                f"{PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR} if the attempts timed out."
             )
         finally:
             os.chdir(original_dir)
@@ -908,7 +917,7 @@ class ProxyExtrasDBManager:
                         # Set migrations directory for Prisma
                         result = subprocess.run(
                             [_get_prisma_command(), "migrate", "deploy"],
-                            timeout=prisma_command_timeout(),
+                            timeout=prisma_migrate_deploy_timeout(),
                             check=True,
                             capture_output=True,
                             text=True,
@@ -1126,7 +1135,11 @@ class ProxyExtrasDBManager:
                     )
                     return True
             except subprocess.TimeoutExpired:
-                logger.info(f"Attempt {attempt + 1} timed out")
+                logger.warning(
+                    "Attempt %s timed out. Raise %s if this database needs longer to apply its schema.",
+                    attempt + 1,
+                    PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR if use_migrate else PRISMA_COMMAND_TIMEOUT_ENV_VAR,
+                )
                 time.sleep(random.randrange(5, 15))
             except subprocess.CalledProcessError as e:
                 attempts_left = 3 - attempt
diff --git a/litellm-rust/AGENTS.md b/litellm-rust/AGENTS.md
index 36a5ad5a8f4..b8b6291283d 100644
--- a/litellm-rust/AGENTS.md
+++ b/litellm-rust/AGENTS.md
@@ -1,6 +1,6 @@
 # AGENTS.md
 
-litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are MODULES inside the layers.
+litellm-rust has four crates. A crate is a layer or shared foundation, not a route. Routes (ocr, realtime, chat) and providers (mistral, openai) are modules inside the layers.
 
 ## Crates
 
@@ -8,9 +8,10 @@ litellm-rust has exactly THREE crates. A crate is a LAYER, not a route. Routes (
 |-------|------|
 | litellm-core | The LiteLLM SDK in Rust. One public entrypoint per top-level call (`messages::messages()`), owning types, transforms, provider resolution, auth, and the provider HTTP call. Call it, get a typed response. |
 | litellm-ai-gateway | The axum server (behind the `server` feature) plus the WebSocket hosts. Translates HTTP/WS to core entrypoints; owns no provider logic and no handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 ## Where a route lives
 
@@ -28,7 +29,7 @@ core/src/messages/
 
 Handlers never live in `ai-gateway`. `ocr`, `audio_transcription`, and `realtime` are still hosted there from before this rule; they move to `core` as they are touched.
 
-Adding a crate: default to a MODULE. New crate ONLY on a real trigger — separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
+Adding a crate: default to a module. A new crate requires a real trigger: separate artifact (binary/cdylib), proc-macro, shared foundation, or publishable standalone. A new provider or route is none of these.
 
 Adding a crate fails crates/core/tests/workspace_crate_allowlist.rs until you update its allowlist and this file — intentional.
 
diff --git a/litellm-rust/CLAUDE.md b/litellm-rust/CLAUDE.md
index fe6ceedbb86..3dcf1853efc 100644
--- a/litellm-rust/CLAUDE.md
+++ b/litellm-rust/CLAUDE.md
@@ -21,12 +21,13 @@ variants of it. The test for a good abstraction is that adding the next provider
 is a few declarative lines, not a new file of duplicated flow. Only diverge from
 the base when behavior is genuinely different, and say so explicitly in the PR.
 
-## Crates (exactly three — see AGENTS.md)
+## Crates (see AGENTS.md)
 
 `litellm-core` **is** the LiteLLM SDK in Rust: it makes the LLM call.
 `litellm-ai-gateway` is an HTTP/WebSocket server in front of it, and
-`litellm-python-bridge` exposes it to the Python SDK. A crate is a **layer**, not
-a route — add modules, not crates.
+`litellm-python-bridge` exposes it to the Python SDK. `litellm-python-interop`
+holds domain-neutral PyO3 primitives shared by Python-facing Rust code. A crate
+is a layer or shared foundation, not a route; add modules, not crates.
 
 ## Core Boundary
 
@@ -175,7 +176,7 @@ cd litellm-rust
 cargo fmt --check
 # the ai-gateway binary + server code is behind the `server` feature
 cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
-cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
+cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
 cargo test --workspace
 ```
 
diff --git a/litellm-rust/Cargo.lock b/litellm-rust/Cargo.lock
index 4388e561026..dd41cf0e84b 100644
--- a/litellm-rust/Cargo.lock
+++ b/litellm-rust/Cargo.lock
@@ -919,6 +919,12 @@ version = "0.3.33"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "b231ed28831efb4a61a08580c4bc233ec56bc009f4cd8f52da2c3cb97df0c109"
 
+[[package]]
+name = "futures-timer"
+version = "3.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "af43fadb8a98512d547e37b4e92e0ced13e205c061b87b4623eff01d918d6968"
+
 [[package]]
 name = "futures-util"
 version = "0.3.33"
@@ -972,6 +978,12 @@ dependencies = [
  "wasm-bindgen",
 ]
 
+[[package]]
+name = "glob"
+version = "0.3.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e4eba85ea1d0a966a983acd07deee566e67395d2d96b6fb39e62b5a833f1eb0b"
+
 [[package]]
 name = "h2"
 version = "0.3.27"
@@ -1432,14 +1444,24 @@ dependencies = [
  "criterion",
  "litellm-ai-gateway",
  "litellm-core",
+ "litellm-python-interop",
  "pyo3",
  "pyo3-async-runtimes",
- "pythonize",
- "serde",
  "serde_json",
  "tokio",
 ]
 
+[[package]]
+name = "litellm-python-interop"
+version = "0.1.0"
+dependencies = [
+ "pyo3",
+ "pythonize",
+ "rstest",
+ "serde",
+ "serde_json",
+]
+
 [[package]]
 name = "litemap"
 version = "0.8.2"
@@ -1627,6 +1649,15 @@ dependencies = [
  "zerocopy",
 ]
 
+[[package]]
+name = "proc-macro-crate"
+version = "3.5.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "e67ba7e9b2b56446f1d419b1d807906278ffa1a658a8a5d8a39dcb1f5a78614f"
+dependencies = [
+ "toml_edit",
+]
+
 [[package]]
 name = "proc-macro2"
 version = "1.0.107"
@@ -1899,6 +1930,12 @@ version = "0.8.11"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d6f6ff9a378485b298a5286656da665ba74413d36db0979633275d2e708145d4"
 
+[[package]]
+name = "relative-path"
+version = "1.9.3"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "ba39f3699c378cd8970968dcbff9c43159ea4cfbd88d43c00b22f2ef10a435d2"
+
 [[package]]
 name = "reqwest"
 version = "0.12.28"
@@ -1956,6 +1993,35 @@ dependencies = [
  "windows-sys 0.52.0",
 ]
 
+[[package]]
+name = "rstest"
+version = "0.26.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "f5a3193c063baaa2a95a33f03035c8a72b83d97a54916055ba22d35ed3839d49"
+dependencies = [
+ "futures-timer",
+ "futures-util",
+ "rstest_macros",
+]
+
+[[package]]
+name = "rstest_macros"
+version = "0.26.1"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9c845311f0ff7951c5506121a9ad75aec44d083c31583b2ea5a30bcb0b0abba0"
+dependencies = [
+ "cfg-if",
+ "glob",
+ "proc-macro-crate",
+ "proc-macro2",
+ "quote",
+ "regex",
+ "relative-path",
+ "rustc_version",
+ "syn 2.0.119",
+ "unicode-ident",
+]
+
 [[package]]
 name = "rustc-hash"
 version = "2.1.3"
@@ -2488,6 +2554,36 @@ dependencies = [
  "tokio",
 ]
 
+[[package]]
+name = "toml_datetime"
+version = "1.1.1+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "3165f65f62e28e0115a00b2ebdd37eb6f3b641855f9d636d3cd4103767159ad7"
+dependencies = [
+ "serde_core",
+]
+
+[[package]]
+name = "toml_edit"
+version = "0.25.13+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "6975367e4d2ef766d86af01ffad14b622fecc8d4357a998fbc4deb6e9bacaf9b"
+dependencies = [
+ "indexmap",
+ "toml_datetime",
+ "toml_parser",
+ "winnow",
+]
+
+[[package]]
+name = "toml_parser"
+version = "1.1.3+spec-1.1.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "1d38ac1cf9b95face32296c0a3ede1fdc270627c9d9c02a7274dd6d960dc4d56"
+dependencies = [
+ "winnow",
+]
+
 [[package]]
 name = "tower"
 version = "0.5.3"
@@ -2903,6 +2999,15 @@ version = "0.52.6"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "589f6da84c646204747d1270a2a5661ea66ed1cced2631d546fdfb155959f9ec"
 
+[[package]]
+name = "winnow"
+version = "1.0.4"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "23b97319f7b8343df12cc98938e5c3eb436064524c8d2b4e30a1d3a36eecdf81"
+dependencies = [
+ "memchr",
+]
+
 [[package]]
 name = "writeable"
 version = "0.6.3"
diff --git a/litellm-rust/Cargo.toml b/litellm-rust/Cargo.toml
index c17a0605fc7..c447d915abe 100644
--- a/litellm-rust/Cargo.toml
+++ b/litellm-rust/Cargo.toml
@@ -2,6 +2,7 @@
 members = [
     "crates/core",
     "crates/ai-gateway",
+    "crates/python-interop",
     "crates/python-bridge",
 ]
 resolver = "2"
@@ -15,12 +16,14 @@ repository = "https://github.com/BerriAI/litellm"
 [workspace.dependencies]
 litellm-core = { path = "crates/core" }
 litellm-ai-gateway = { path = "crates/ai-gateway", default-features = false }
+litellm-python-interop = { path = "crates/python-interop" }
 axum = "0.7"
 pyo3 = "0.29.0"
 pyo3-async-runtimes = { version = "0.29.0", features = ["tokio-runtime"] }
 pythonize = "0.29.0"
 rand = "0.8"
 reqwest = { version = "0.12", default-features = false, features = ["blocking", "json", "rustls-tls", "http2", "stream"] }
+rstest = "0.26.1"
 serde = { version = "1.0", features = ["derive"] }
 serde_json = "1.0"
 sha2 = "0.10"
diff --git a/litellm-rust/README.md b/litellm-rust/README.md
index bcccf93300b..a0d79c6f0a5 100644
--- a/litellm-rust/README.md
+++ b/litellm-rust/README.md
@@ -26,9 +26,10 @@ coverage and production evidence.
 |-------|------|
 | litellm-core | The SDK. Per-route entrypoints (`messages::messages()`), types, provider transforms (modules under `providers/`), provider resolution, auth, the provider HTTP call, and the router. |
 | litellm-ai-gateway | The axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. Owns API registration, domain wiring, and Python exception mapping. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 ## Layout
 
@@ -38,7 +39,8 @@ crates/
     src/messages/   mod.rs (entrypoint), types, transformation, prepare, handler, client
     src/providers/anthropic/messages/transformation.rs
   ai-gateway/     Axum server + WebSocket hosts; calls core entrypoints.
-  python-bridge/  PyO3 bridge for Python LiteLLM.
+  python-interop/ Domain-neutral PyO3 conversion and GIL primitives.
+  python-bridge/  PyO3 API adapter for Python LiteLLM.
 ```
 
 The folder shape follows the Python provider tree:
diff --git a/litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md b/litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md
index a1860d8a9c9..4a689cb9579 100644
--- a/litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md
+++ b/litellm-rust/crates/CODING_STANDARDS/PROVIDER_CODING_STANDARDS.md
@@ -54,6 +54,6 @@ Rules for adding or changing an LLM provider/route in `litellm-rust`. `messages`
     cd litellm-rust
     cargo fmt --check
     cargo clippy -p litellm-ai-gateway --all-targets --features server -- -D warnings
-    cargo clippy -p litellm-core -p litellm-python-bridge --all-targets -- -D warnings
+    cargo clippy -p litellm-core -p litellm-python-interop -p litellm-python-bridge --all-targets -- -D warnings
     cargo test --workspace
     ```
diff --git a/litellm-rust/crates/ai-gateway/README.md b/litellm-rust/crates/ai-gateway/README.md
index 7a6c620ee84..5cbb47220be 100644
--- a/litellm-rust/crates/ai-gateway/README.md
+++ b/litellm-rust/crates/ai-gateway/README.md
@@ -6,15 +6,16 @@ dials OpenAI upstream, and splices the two sockets frame-by-frame.
 
 ## Crates
 
-`litellm-rust` is exactly three crates (a crate is a **layer**, not a route):
+`litellm-rust` has four crates. A crate is a layer or shared foundation, not a route:
 
 | Crate | Role |
 |-------|------|
 | litellm-core | The LiteLLM SDK in Rust — per-route entrypoints (`messages::messages()`) that resolve the provider, transform, and make the call; plus types, provider transforms, and the router. |
 | litellm-ai-gateway | The Axum server (behind the `server` feature) and WebSocket hosts. Translates HTTP/WS to core entrypoints; no provider handlers. |
-| litellm-python-bridge | PyO3 cdylib exposing Rust to the litellm Python SDK — marshals Python objects and calls core entrypoints. |
+| litellm-python-interop | Domain-neutral PyO3 foundation for GIL handling and typed Python/Serde conversion. |
+| litellm-python-bridge | PyO3 cdylib exposing LiteLLM Rust APIs to the Python SDK. |
 
-Dependency direction (acyclic): litellm-core ← litellm-ai-gateway ← litellm-python-bridge.
+Dependency direction is acyclic: `litellm-python-bridge` depends on the domain layers and `litellm-python-interop`; the interop foundation depends on no LiteLLM domain crate.
 
 - **Client endpoint:** `wss://<host>/v1/realtime?model=<model>` (WebSocket)
 - **Auth:** `Authorization: Bearer $LITELLM_MASTER_KEY` (fails closed if unset)
diff --git a/litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs b/litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs
index 270d5c2d97a..140bc8aeea8 100644
--- a/litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs
+++ b/litellm-rust/crates/ai-gateway/src/audio_transcription/common_utils.rs
@@ -1,10 +1,8 @@
-use std::collections::BTreeMap;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionProviderConfig;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::providers::bedrock::audio_transcription::BEDROCK_AUDIO_TRANSCRIPTION_CONFIG;
 use serde_json::{Map, Value};
+use std::collections::BTreeMap;
 
 pub(super) fn audio_transcription_provider_config(
     provider: &str,
@@ -17,7 +15,7 @@ pub(super) fn audio_transcription_provider_config(
 
 pub(super) fn string_headers(
     headers: Option<Map<String, Value>>,
-) -> CoreResult<BTreeMap<String, String>> {
+) -> Result<BTreeMap<String, String>, Error> {
     headers
         .unwrap_or_default()
         .into_iter()
@@ -26,7 +24,7 @@ pub(super) fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "audio transcription extra_headers.{key} must be a string"
                     ))
                 })
diff --git a/litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs b/litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs
index 33c13550f58..1bdd4ae72a2 100644
--- a/litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs
+++ b/litellm-rust/crates/ai-gateway/src/audio_transcription/handler.rs
@@ -1,11 +1,9 @@
-use std::time::SystemTime;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::providers::bedrock::audio_transcription::aws_auth_config;
 use litellm_core::providers::bedrock::aws_base::{resolve_credentials, sign_bedrock_post};
 use serde_json::Value;
+use std::time::SystemTime;
 
 use super::common_utils::truncate_error_body;
 use super::types::ProviderAudioTranscriptionRequest;
@@ -13,10 +11,9 @@ use crate::client::http_client;
 
 pub(crate) async fn execute_audio_transcription_provider_call(
     request: ProviderAudioTranscriptionRequest,
-) -> CoreResult<Value> {
-    let body = serde_json::to_vec(&request.body).map_err(|error| {
-        CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
-    })?;
+) -> Result<Value, Error> {
+    let body = serde_json::to_vec(&request.body)
+        .map_err(|error| Error::InvalidRequest(format!("invalid audio request body: {error}")))?;
     let mut request_builder = http_client().post(&request.url).body(body.clone());
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -27,21 +24,20 @@ pub(crate) async fn execute_audio_transcription_provider_call(
     let response = request_builder
         .send()
         .await
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     let status = response.status();
     let text = response
         .text()
         .await
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
-    let response_json: Value = serde_json::from_str(&text).map_err(|error| {
-        CoreError::InvalidResponse(format!("invalid audio response JSON: {error}"))
-    })?;
+    let response_json: Value = serde_json::from_str(&text)
+        .map_err(|error| Error::InvalidResponse(format!("invalid audio response JSON: {error}")))?;
     Ok(request
         .config
         .transform_transcription_response(&request.model, response_json)?
@@ -51,14 +47,13 @@ pub(crate) async fn execute_audio_transcription_provider_call(
 pub(crate) async fn sign_request(
     request: &ProviderAudioTranscriptionRequest,
     optional_params: &serde_json::Map<String, Value>,
-) -> CoreResult<ProviderAudioTranscriptionRequest> {
+) -> Result<ProviderAudioTranscriptionRequest, Error> {
     let env_lookup = environment_lookup;
     let auth = request
         .config
         .auth_strategy(&request.model, optional_params, &env_lookup)?;
-    let body = serde_json::to_vec(&request.body).map_err(|error| {
-        CoreError::InvalidRequest(format!("invalid audio request body: {error}"))
-    })?;
+    let body = serde_json::to_vec(&request.body)
+        .map_err(|error| Error::InvalidRequest(format!("invalid audio request body: {error}")))?;
     let mut headers = super::common_utils::string_headers(None)?;
     headers.insert("Content-Type".to_string(), "application/json".to_string());
     headers.extend(request.upstream_headers.iter().cloned());
diff --git a/litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs b/litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs
index 0c9faeda6e7..5e1240de759 100644
--- a/litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs
+++ b/litellm-rust/crates/ai-gateway/src/audio_transcription/hooks.rs
@@ -1,11 +1,9 @@
-use std::future::Future;
-use std::pin::Pin;
-
-use litellm_core::CoreResult;
 use litellm_core::audio_transcription::transformation::AudioTranscriptionAuth;
 use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use serde_json::{Map, Value, json};
+use std::future::Future;
+use std::pin::Pin;
 
 use super::common_utils::{audio_transcription_provider_config, has_header, string_headers};
 use super::handler::sign_request;
@@ -26,7 +24,7 @@ pub(crate) struct AudioTranscriptionLifecycleHooks {
     request_metadata: RequestMetadata,
 }
 
-type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type AudioFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 type AudioLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
 
 impl AudioTranscriptionLifecycleHooks {
@@ -45,7 +43,7 @@ impl AudioTranscriptionLifecycleHooks {
     async fn run_pre_call_guardrails(
         &self,
         request: PreparedAudioTranscriptionRequest,
-    ) -> CoreResult<PreparedAudioTranscriptionRequest> {
+    ) -> Result<PreparedAudioTranscriptionRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -63,17 +61,17 @@ impl AudioTranscriptionLifecycleHooks {
             .await
             .map_err(guardrail_error_to_core_error)?;
         let Value::Object(mut data) = guardrail_request.data else {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "audio transcription pre_call guardrail must return an object".to_string(),
             ));
         };
         let audio = data.remove("audio").ok_or_else(|| {
-            CoreError::InvalidRequest("audio transcription guardrail removed audio".to_string())
+            Error::InvalidRequest("audio transcription guardrail removed audio".to_string())
         })?;
         let optional_params = match data.remove("optional_params") {
             Some(Value::Object(value)) => value,
             Some(_) => {
-                return Err(CoreError::InvalidRequest(
+                return Err(Error::InvalidRequest(
                     "audio transcription optional_params must be an object".to_string(),
                 ));
             }
@@ -89,9 +87,9 @@ impl AudioTranscriptionLifecycleHooks {
     async fn prepare_provider_request(
         &self,
         request: PreparedAudioTranscriptionRequest,
-    ) -> CoreResult<ProviderAudioTranscriptionRequest> {
+    ) -> Result<ProviderAudioTranscriptionRequest, Error> {
         let config = audio_transcription_provider_config(&request.custom_llm_provider)
-            .ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
+            .ok_or_else(|| Error::InvalidProvider(request.custom_llm_provider.clone()))?;
         let env_lookup = super::handler::environment_lookup;
         let headers = string_headers(request.extra_headers)?;
         let url = config.complete_url(
@@ -135,7 +133,7 @@ impl AudioTranscriptionLifecycleHooks {
     async fn run_during_call_guardrails(
         &self,
         request: ProviderAudioTranscriptionRequest,
-    ) -> CoreResult<ProviderAudioTranscriptionRequest> {
+    ) -> Result<ProviderAudioTranscriptionRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -153,12 +151,12 @@ impl AudioTranscriptionLifecycleHooks {
             .await
             .map_err(guardrail_error_to_core_error)?;
         let Value::Object(mut data) = guardrail_request.data else {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "audio transcription during_call guardrail must return an object".to_string(),
             ));
         };
         let body = data.remove("body").ok_or_else(|| {
-            CoreError::InvalidRequest("audio transcription guardrail removed body".to_string())
+            Error::InvalidRequest("audio transcription guardrail removed body".to_string())
         })?;
         Ok(ProviderAudioTranscriptionRequest { body, ..request })
     }
@@ -241,7 +239,7 @@ impl CallLifecycleHooks<PreparedAudioTranscriptionRequest, ProviderAudioTranscri
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -281,22 +279,22 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
     }
 }
 
-fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
-    CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
+fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
+    Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
 }
 
-fn core_error_kind(error: &CoreError) -> &'static str {
+fn core_error_kind(error: &Error) -> &'static str {
     match error {
-        CoreError::Auth(_) => "AuthError",
-        CoreError::InvalidProvider(_) => "InvalidProvider",
-        CoreError::InvalidRequest(_) => "InvalidRequest",
-        CoreError::InvalidType { .. } => "InvalidType",
-        CoreError::MissingField(_) => "MissingField",
-        CoreError::Http { .. } => "HttpError",
-        CoreError::InvalidResponse(_) => "InvalidResponse",
-        CoreError::Network(_) => "NetworkError",
-        CoreError::Connect(_) => "ConnectError",
-        CoreError::Routing(_) => "RoutingError",
-        CoreError::Unsupported(_) => "UnsupportedRequest",
+        Error::Auth(_) => "AuthError",
+        Error::InvalidProvider(_) => "InvalidProvider",
+        Error::InvalidRequest(_) => "InvalidRequest",
+        Error::InvalidType { .. } => "InvalidType",
+        Error::MissingField(_) => "MissingField",
+        Error::Http { .. } => "HttpError",
+        Error::InvalidResponse(_) => "InvalidResponse",
+        Error::Network(_) => "NetworkError",
+        Error::Connect(_) => "ConnectError",
+        Error::Routing(_) => "RoutingError",
+        Error::Unsupported(_) => "UnsupportedRequest",
     }
 }
diff --git a/litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs b/litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs
index 5d33d912c40..3983846d7b6 100644
--- a/litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs
+++ b/litellm-rust/crates/ai-gateway/src/audio_transcription/mod.rs
@@ -1,4 +1,4 @@
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::call_lifecycle::CallLifecycle;
 use serde_json::Value;
 
@@ -13,7 +13,7 @@ pub use types::AudioTranscriptionRequest;
 use handler::execute_audio_transcription_provider_call;
 use prepare::{PreparedAudioTranscriptionCall, prepare_audio_transcription_call};
 
-pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> CoreResult<Value> {
+pub async fn audio_transcription(request: AudioTranscriptionRequest<'_>) -> Result<Value, Error> {
     let PreparedAudioTranscriptionCall { request, hooks } =
         prepare_audio_transcription_call(request);
     CallLifecycle::default()
diff --git a/litellm-rust/crates/ai-gateway/src/io/realtime.rs b/litellm-rust/crates/ai-gateway/src/io/realtime.rs
index 845e7bf9527..662f7328982 100644
--- a/litellm-rust/crates/ai-gateway/src/io/realtime.rs
+++ b/litellm-rust/crates/ai-gateway/src/io/realtime.rs
@@ -15,8 +15,7 @@ use std::time::Duration;
 
 use futures_util::stream::{SplitSink, SplitStream};
 use futures_util::{Sink, SinkExt, Stream, StreamExt};
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::realtime::transformation::RealtimeProviderConfig;
 use litellm_core::realtime::types::RealtimeEvent;
 use tokio::net::TcpStream;
@@ -48,7 +47,7 @@ pub(crate) type UpstreamRx = SplitStream<UpstreamWs>;
 /// Resolve the OpenAI API key from the explicit param or the environment.
 ///
 /// Blank/whitespace values are treated as absent (guard at resolution time).
-pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
+pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
@@ -58,7 +57,7 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
                 .ok()
                 .filter(|key| !key.trim().is_empty())
         })
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 /// Open the upstream WebSocket to OpenAI for `(model, api_key, api_base)`.
@@ -70,24 +69,24 @@ pub(crate) async fn dial_upstream(
     model: &str,
     api_key: &str,
     api_base: Option<&str>,
-) -> CoreResult<UpstreamWs> {
+) -> Result<UpstreamWs, Error> {
     let url = OPENAI_REALTIME_CONFIG.complete_url(api_base, model);
 
     let mut request = url
         .as_str()
         .into_client_request()
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     // GA realtime: only Authorization. The legacy OpenAI-Beta header triggers
     // beta_api_shape_disabled, so we do not send it.
     request.headers_mut().insert(
         AUTHORIZATION,
         HeaderValue::from_str(&format!("Bearer {api_key}"))
-            .map_err(|err| CoreError::Auth(err.to_string()))?,
+            .map_err(|err| Error::Auth(err.to_string()))?,
     );
 
     let (upstream, _response) = connect_async(request)
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     Ok(upstream)
 }
 
@@ -96,22 +95,22 @@ pub(crate) async fn dial_upstream(
 /// Used by the pool to pre-read OpenAI's unprompted `session.created`. Returns an
 /// error on a non-text frame, a closed socket, or undecodable JSON so the pool can
 /// discard a misbehaving socket rather than warm it.
-pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> CoreResult<RealtimeEvent> {
+pub(crate) async fn read_event(upstream_rx: &mut UpstreamRx) -> Result<RealtimeEvent, Error> {
     loop {
         let message = upstream_rx
             .next()
             .await
-            .ok_or_else(|| CoreError::Network("upstream closed before first event".to_string()))?
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .ok_or_else(|| Error::Network("upstream closed before first event".to_string()))?
+            .map_err(|err| Error::Network(err.to_string()))?;
         match message {
             Message::Text(text) => {
                 return serde_json::from_str(&text)
-                    .map_err(|err| CoreError::InvalidResponse(err.to_string()));
+                    .map_err(|err| Error::InvalidResponse(err.to_string()));
             }
             // Ignore protocol frames (ping/pong) while waiting for the first event.
             Message::Ping(_) | Message::Pong(_) => continue,
             Message::Close(_) => {
-                return Err(CoreError::Network(
+                return Err(Error::Network(
                     "upstream closed before first event".to_string(),
                 ));
             }
@@ -139,7 +138,7 @@ pub(crate) async fn splice<In, Out>(
     mut observe: impl FnMut(&RealtimeEvent) + Send,
     mut client_in: In,
     mut client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
@@ -154,7 +153,7 @@ where
             client_out
                 .send(outbound)
                 .await
-                .map_err(|err| CoreError::Network(err.to_string()))?;
+                .map_err(|err| Error::Network(err.to_string()))?;
         }
     }
 
@@ -175,26 +174,26 @@ where
                 // inflate its own spend log. Logging observes upstream events only.
                 for outbound in config.transform_realtime_request(&event, model)?.events {
                     let payload = serde_json::to_string(&outbound)
-                        .map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
+                        .map_err(|err| Error::InvalidResponse(err.to_string()))?;
                     upstream_tx
                         .send(Message::Text(payload))
                         .await
-                        .map_err(|err| CoreError::Network(err.to_string()))?;
+                        .map_err(|err| Error::Network(err.to_string()))?;
                 }
             }
             // upstream -> client
             upstream_message = upstream_rx.next() => {
                 let Some(message) = upstream_message else { break }; // upstream closed
-                match message.map_err(|err| CoreError::Network(err.to_string()))? {
+                match message.map_err(|err| Error::Network(err.to_string()))? {
                     Message::Text(text) => {
                         let event: RealtimeEvent = serde_json::from_str(&text)
-                            .map_err(|err| CoreError::InvalidResponse(err.to_string()))?;
+                            .map_err(|err| Error::InvalidResponse(err.to_string()))?;
                         observe(&event);
                         for outbound in config.transform_realtime_response(&event, model)?.events {
                             client_out
                                 .send(outbound)
                                 .await
-                                .map_err(|err| CoreError::Network(err.to_string()))?;
+                                .map_err(|err| Error::Network(err.to_string()))?;
                         }
                     }
                     Message::Close(_) => break,
@@ -225,7 +224,7 @@ pub async fn realtime<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
@@ -258,7 +257,7 @@ pub async fn realtime_warm<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
diff --git a/litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs b/litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs
index 4a1a3cd1166..49e9c459a88 100644
--- a/litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs
+++ b/litellm-rust/crates/ai-gateway/src/io/realtime_pool.rs
@@ -28,7 +28,7 @@ use std::sync::{Arc, Mutex};
 use std::time::{Duration, Instant};
 
 use futures_util::StreamExt;
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::realtime::types::RealtimeEvent;
 
 use crate::io::realtime::{
@@ -438,7 +438,7 @@ impl RealtimePool {
 ///
 /// `key.api_key` is already resolved (non-blank). The first frame OpenAI sends
 /// unprompted is `session.created`; we buffer exactly that and read nothing more.
-async fn warm_one(key: &UpstreamKey) -> CoreResult<WarmConnection> {
+async fn warm_one(key: &UpstreamKey) -> Result<WarmConnection, Error> {
     let upstream: UpstreamWs =
         dial_upstream(&key.model, &key.api_key, key.api_base.as_deref()).await?;
     let (tx, mut rx) = upstream.split();
diff --git a/litellm-rust/crates/ai-gateway/src/io/responses_ws.rs b/litellm-rust/crates/ai-gateway/src/io/responses_ws.rs
index 9b51019f4bc..0b01747b1a5 100644
--- a/litellm-rust/crates/ai-gateway/src/io/responses_ws.rs
+++ b/litellm-rust/crates/ai-gateway/src/io/responses_ws.rs
@@ -4,10 +4,10 @@ use std::time::Duration;
 
 use futures_util::stream::{SplitSink, SplitStream};
 use futures_util::{Sink, SinkExt, Stream, StreamExt};
+use litellm_core::Error;
 use litellm_core::providers::openai::responses::transformation::OPENAI_RESPONSES_WS_CONFIG;
 use litellm_core::responses::types::ResponsesWsEvent;
 use litellm_core::responses::websocket::ResponsesWebSocketProviderConfig;
-use litellm_core::{CoreError, CoreResult};
 use tokio::net::TcpStream;
 use tokio::sync::Mutex;
 use tokio_tungstenite::tungstenite::Message;
@@ -37,51 +37,49 @@ impl ResponsesWebSocketConnection {
         url: &str,
         headers: &HashMap<String, String>,
         timeout: Option<Duration>,
-    ) -> CoreResult<Self> {
+    ) -> Result<Self, Error> {
         let mut request = url
             .into_client_request()
-            .map_err(|error| CoreError::Network(error.to_string()))?;
+            .map_err(|error| Error::Network(error.to_string()))?;
         for (name, value) in headers {
             let header_name = name
                 .parse::<HeaderName>()
-                .map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
+                .map_err(|error| Error::InvalidRequest(error.to_string()))?;
             let header_value = HeaderValue::from_str(value)
-                .map_err(|error| CoreError::InvalidRequest(error.to_string()))?;
+                .map_err(|error| Error::InvalidRequest(error.to_string()))?;
             request.headers_mut().insert(header_name, header_value);
         }
         let connect = connect_async(request);
         let result = match timeout {
             Some(timeout) => tokio::time::timeout(timeout, connect).await.map_err(|_| {
-                CoreError::Network("Responses WebSocket connection timed out".to_string())
+                Error::Network("Responses WebSocket connection timed out".to_string())
             })?,
             None => connect.await,
         };
         let (socket, _) = result.map_err(|error| match error {
-            tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
+            tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
                 status: response.status().as_u16(),
                 body: String::new(),
             },
-            other => CoreError::Network(other.to_string()),
+            other => Error::Network(other.to_string()),
         })?;
         Ok(Self {
             socket: Arc::new(Mutex::new(Some(socket))),
         })
     }
 
-    pub async fn send_text(&self, text: String) -> CoreResult<()> {
+    pub async fn send_text(&self, text: String) -> Result<(), Error> {
         let mut socket = self.socket.lock().await;
         let Some(socket) = socket.as_mut() else {
-            return Err(CoreError::Network(
-                "Responses WebSocket is closed".to_string(),
-            ));
+            return Err(Error::Network("Responses WebSocket is closed".to_string()));
         };
         socket
             .send(Message::Text(text))
             .await
-            .map_err(|error| CoreError::Network(error.to_string()))
+            .map_err(|error| Error::Network(error.to_string()))
     }
 
-    pub async fn recv_text(&self) -> CoreResult<Option<String>> {
+    pub async fn recv_text(&self) -> Result<Option<String>, Error> {
         let mut socket_guard = self.socket.lock().await;
         let Some(socket) = socket_guard.as_mut() else {
             return Ok(None);
@@ -90,27 +88,27 @@ impl ResponsesWebSocketConnection {
             Some(Ok(Message::Text(text))) => Ok(Some(text)),
             Some(Ok(Message::Binary(bytes))) => String::from_utf8(bytes.to_vec())
                 .map(Some)
-                .map_err(|error| CoreError::InvalidResponse(error.to_string())),
+                .map_err(|error| Error::InvalidResponse(error.to_string())),
             Some(Ok(Message::Close(_))) | None => Ok(None),
             Some(Ok(_)) => Ok(None),
-            Some(Err(error)) => Err(CoreError::Network(error.to_string())),
+            Some(Err(error)) => Err(Error::Network(error.to_string())),
         }
     }
 
-    pub async fn close(&self) -> CoreResult<()> {
+    pub async fn close(&self) -> Result<(), Error> {
         let mut socket = self.socket.lock().await;
         if let Some(socket) = socket.as_mut() {
             socket
                 .close(None)
                 .await
-                .map_err(|error| CoreError::Network(error.to_string()))?;
+                .map_err(|error| Error::Network(error.to_string()))?;
         }
         *socket = None;
         Ok(())
     }
 }
 
-pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
+pub(crate) fn resolve_api_key(api_key: Option<&str>) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|value| !value.is_empty())
@@ -120,38 +118,38 @@ pub(crate) fn resolve_api_key(api_key: Option<&str>) -> CoreResult<String> {
                 .ok()
                 .filter(|value| !value.trim().is_empty())
         })
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 async fn dial_upstream(
     model: &str,
     api_key: &str,
     api_base: Option<&str>,
-) -> CoreResult<ResponsesUpstreamWs> {
+) -> Result<ResponsesUpstreamWs, Error> {
     let url = OPENAI_RESPONSES_WS_CONFIG.complete_websocket_url(api_base, model);
     let mut request = url
         .as_str()
         .into_client_request()
-        .map_err(|error| CoreError::Network(error.to_string()))?;
+        .map_err(|error| Error::Network(error.to_string()))?;
     request.headers_mut().insert(
         AUTHORIZATION,
         HeaderValue::from_str(&format!("Bearer {api_key}"))
-            .map_err(|error| CoreError::Auth(error.to_string()))?,
+            .map_err(|error| Error::Auth(error.to_string()))?,
     );
     let result = tokio::time::timeout(
         Duration::from_secs(DEFAULT_RESPONSES_WS_CONNECT_TIMEOUT_SECS),
         connect_async(request),
     )
     .await
-    .map_err(|_| CoreError::Network("Responses WebSocket connection timed out".to_string()))?;
+    .map_err(|_| Error::Network("Responses WebSocket connection timed out".to_string()))?;
     result
         .map(|(socket, _)| socket)
         .map_err(|error| match error {
-            tokio_tungstenite::tungstenite::Error::Http(response) => CoreError::Http {
+            tokio_tungstenite::tungstenite::Error::Http(response) => Error::Http {
                 status: response.status().as_u16(),
                 body: String::new(),
             },
-            other => CoreError::Network(other.to_string()),
+            other => Error::Network(other.to_string()),
         })
 }
 
@@ -166,7 +164,7 @@ impl ResponsesWebSocketStreaming {
         observe: impl FnMut(&ResponsesWsEvent) + Send,
         client_in: In,
         client_out: Out,
-    ) -> CoreResult<()>
+    ) -> Result<(), Error>
     where
         In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
         Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -193,7 +191,7 @@ pub(crate) async fn splice<In, Out>(
     mut observe: impl FnMut(&ResponsesWsEvent) + Send,
     mut client_in: In,
     mut client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -210,18 +208,18 @@ where
                     .events
                 {
                     let payload = serde_json::to_string(&outbound)
-                        .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                        .map_err(|error| Error::InvalidResponse(error.to_string()))?;
                     upstream_tx.send(Message::Text(payload))
                         .await
-                        .map_err(|error| CoreError::Network(error.to_string()))?;
+                        .map_err(|error| Error::Network(error.to_string()))?;
                 }
             }
             message = upstream_rx.next() => {
                 let Some(message) = message else { break };
-                match message.map_err(|error| CoreError::Network(error.to_string()))? {
+                match message.map_err(|error| Error::Network(error.to_string()))? {
                     Message::Text(text) => {
                         let event = serde_json::from_str::<ResponsesWsEvent>(&text)
-                            .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                            .map_err(|error| Error::InvalidResponse(error.to_string()))?;
                         observe(&event);
                         for outbound in OPENAI_RESPONSES_WS_CONFIG
                             .transform_ws_response(&event, model)?
@@ -229,7 +227,7 @@ where
                         {
                             client_out.send(outbound)
                                 .await
-                                .map_err(|error| CoreError::Network(error.to_string()))?;
+                                .map_err(|error| Error::Network(error.to_string()))?;
                         }
                     }
                     Message::Close(_) => break,
@@ -252,7 +250,7 @@ pub async fn async_responses_websocket<In, Out>(
     mut observe: impl FnMut(&ResponsesWsEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -267,11 +265,11 @@ where
             .events
         {
             let payload = serde_json::to_string(&outbound)
-                .map_err(|error| CoreError::InvalidResponse(error.to_string()))?;
+                .map_err(|error| Error::InvalidResponse(error.to_string()))?;
             upstream_tx
                 .send(Message::Text(payload))
                 .await
-                .map_err(|error| CoreError::Network(error.to_string()))?;
+                .map_err(|error| Error::Network(error.to_string()))?;
         }
     }
     ResponsesWebSocketStreaming::bidirectional_forward(
@@ -296,7 +294,7 @@ pub async fn responses_ws<In, Out>(
     observe: impl FnMut(&ResponsesWsEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
@@ -514,7 +512,7 @@ mod tests {
         )
         .await
         .expect_err("status error");
-        assert!(matches!(error, CoreError::Http { status: 401, .. }));
+        assert!(matches!(error, Error::Http { status: 401, .. }));
         server.await.expect("server task");
     }
 
@@ -543,7 +541,7 @@ mod tests {
         )
         .await
         .expect_err("status error");
-        assert!(matches!(error, CoreError::Http { status: 500, .. }));
+        assert!(matches!(error, Error::Http { status: 500, .. }));
         server.await.expect("server task");
     }
 }
diff --git a/litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs b/litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs
index 9bc2818b6e7..e0ce165dc93 100644
--- a/litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs
+++ b/litellm-rust/crates/ai-gateway/src/ocr/common_utils.rs
@@ -3,8 +3,7 @@ use std::time::{Duration, Instant};
 
 use base64::Engine;
 use base64::engine::general_purpose::STANDARD as BASE64_STANDARD;
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrProviderConfig;
 use reqwest::Url;
 use serde_json::{Map, Value};
@@ -56,7 +55,7 @@ fn is_azure_document_intelligence_model(model: &str) -> bool {
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     extra_headers
         .unwrap_or_default()
         .into_iter()
@@ -65,7 +64,7 @@ pub(super) fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "OCR extra_headers.{key} must be a string, got {}",
                         litellm_core::error::json_type_name(&value)
                     ))
@@ -80,7 +79,7 @@ pub(super) fn has_header(headers: &[(String, String)], name: &str) -> bool {
         .any(|(key, _)| key.eq_ignore_ascii_case(name))
 }
 
-fn document_url_field(document: &Value) -> CoreResult<Option<(&str, &str)>> {
+fn document_url_field(document: &Value) -> Result<Option<(&str, &str)>, Error> {
     let Some(object) = document.as_object() else {
         return Ok(None);
     };
@@ -138,13 +137,13 @@ fn is_blocked_ip(ip: IpAddr) -> bool {
     }
 }
 
-fn blocked_url_error(url: &Url) -> CoreError {
-    CoreError::InvalidRequest(format!(
+fn blocked_url_error(url: &Url) -> Error {
+    Error::InvalidRequest(format!(
         "OCR document URL rejected by SSRF protection: {url}"
     ))
 }
 
-async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
+async fn validate_safe_fetch_url(url: &Url) -> Result<(), Error> {
     if !matches!(url.scheme(), "http" | "https") {
         return Err(blocked_url_error(url));
     }
@@ -162,7 +161,7 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
         .ok_or_else(|| blocked_url_error(url))?;
     let addresses = tokio::net::lookup_host((host, port))
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let mut saw_address = false;
     for address in addresses {
         saw_address = true;
@@ -176,25 +175,25 @@ async fn validate_safe_fetch_url(url: &Url) -> CoreResult<()> {
     Ok(())
 }
 
-fn redirect_location(response: &reqwest::Response, url: &Url) -> CoreResult<Url> {
+fn redirect_location(response: &reqwest::Response, url: &Url) -> Result<Url, Error> {
     let location = response
         .headers()
         .get(reqwest::header::LOCATION)
         .and_then(|value| value.to_str().ok())
         .ok_or_else(|| {
-            CoreError::InvalidResponse("OCR document redirect missing Location header".to_string())
+            Error::InvalidResponse("OCR document redirect missing Location header".to_string())
         })?;
     url.join(location)
-        .map_err(|err| CoreError::InvalidResponse(format!("invalid OCR document redirect: {err}")))
+        .map_err(|err| Error::InvalidResponse(format!("invalid OCR document redirect: {err}")))
 }
 
-async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)> {
+async fn safe_get_document_url(url: &str) -> Result<(Url, reqwest::Response), Error> {
     let client = reqwest::Client::builder()
         .redirect(reqwest::redirect::Policy::none())
         .build()
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let mut current_url = Url::parse(url)
-        .map_err(|err| CoreError::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
+        .map_err(|err| Error::InvalidRequest(format!("invalid OCR document URL: {err}")))?;
 
     for _ in 0..MAX_SAFE_FETCH_REDIRECTS {
         validate_safe_fetch_url(&current_url).await?;
@@ -202,28 +201,28 @@ async fn safe_get_document_url(url: &str) -> CoreResult<(Url, reqwest::Response)
             .get(current_url.clone())
             .send()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         if !response.status().is_redirection() {
             return Ok((current_url, response));
         }
         current_url = redirect_location(&response, &current_url)?;
     }
 
-    Err(CoreError::InvalidRequest(
+    Err(Error::InvalidRequest(
         "Too many redirects while fetching OCR document URL".to_string(),
     ))
 }
 
-fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> CoreResult<()> {
+fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Result<(), Error> {
     if max_bytes == 0 {
-        return Err(CoreError::InvalidRequest(format!(
+        return Err(Error::InvalidRequest(format!(
             "OCR document URL download is disabled (MAX_IMAGE_URL_DOWNLOAD_SIZE_MB=0). url={url}"
         )));
     }
     if content_length > max_bytes {
         let size_mb = content_length as f64 / (1024.0 * 1024.0);
         let max_size_mb = max_bytes as f64 / (1024.0 * 1024.0);
-        return Err(CoreError::InvalidRequest(format!(
+        return Err(Error::InvalidRequest(format!(
             "OCR document size ({size_mb:.2}MB) exceeds maximum allowed size ({max_size_mb:.2}MB). url={url}"
         )));
     }
@@ -233,7 +232,7 @@ fn enforce_download_size(content_length: u64, max_bytes: u64, url: &Url) -> Core
 async fn read_response_with_limit(
     mut response: reqwest::Response,
     url: &Url,
-) -> CoreResult<Vec<u8>> {
+) -> Result<Vec<u8>, Error> {
     let max_bytes = max_document_download_bytes();
     if let Some(content_length) = response.content_length() {
         enforce_download_size(content_length, max_bytes, url)?;
@@ -246,7 +245,7 @@ async fn read_response_with_limit(
     while let Some(chunk) = response
         .chunk()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?
+        .map_err(|err| Error::Network(err.to_string()))?
     {
         bytes_downloaded += chunk.len() as u64;
         enforce_download_size(bytes_downloaded, max_bytes, url)?;
@@ -255,7 +254,7 @@ async fn read_response_with_limit(
     Ok(bytes)
 }
 
-pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreResult<Value> {
+pub(super) async fn convert_document_url_to_data_uri(document: Value) -> Result<Value, Error> {
     let Some((field, url)) = document_url_field(&document)? else {
         return Ok(document);
     };
@@ -267,7 +266,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
     let status = response.status();
     if !status.is_success() {
         let body = response.text().await.unwrap_or_default();
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&body),
         });
@@ -290,7 +289,7 @@ pub(super) async fn convert_document_url_to_data_uri(document: Value) -> CoreRes
     let mut transformed = document
         .as_object()
         .cloned()
-        .ok_or_else(|| CoreError::InvalidRequest("OCR document must be an object".to_string()))?;
+        .ok_or_else(|| Error::InvalidRequest("OCR document must be an object".to_string()))?;
     transformed.insert(field.to_string(), Value::String(data_uri));
     Ok(Value::Object(transformed))
 }
@@ -316,11 +315,11 @@ fn retry_after_secs(response: &reqwest::Response) -> u64 {
         .unwrap_or(2)
 }
 
-fn operation_status(response_json: &Value) -> CoreResult<&str> {
+fn operation_status(response_json: &Value) -> Result<&str, Error> {
     let status = response_json
         .get("status")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("status"))?;
+        .ok_or(Error::MissingField("status"))?;
     match status {
         "succeeded" => Ok("succeeded"),
         "running" | "notStarted" => Ok("running"),
@@ -330,11 +329,11 @@ fn operation_status(response_json: &Value) -> CoreResult<&str> {
                 .and_then(|error| error.get("message"))
                 .and_then(Value::as_str)
                 .unwrap_or("Unknown error");
-            Err(CoreError::InvalidResponse(format!(
+            Err(Error::InvalidResponse(format!(
                 "Azure Document Intelligence analysis failed: {message}"
             )))
         }
-        other => Err(CoreError::InvalidResponse(format!(
+        other => Err(Error::InvalidResponse(format!(
             "Unknown operation status: {other}"
         ))),
     }
@@ -345,9 +344,9 @@ pub(super) async fn poll_document_intelligence(
     original_url: &str,
     headers: &[(String, String)],
     timeout: Option<Duration>,
-) -> CoreResult<Value> {
+) -> Result<Value, Error> {
     if !same_origin(operation_url, original_url) {
-        return Err(CoreError::InvalidResponse(
+        return Err(Error::InvalidResponse(
             "Azure Document Intelligence: rejected cross-origin polling URL".to_string(),
         ));
     }
@@ -358,7 +357,7 @@ pub(super) async fn poll_document_intelligence(
     ));
     loop {
         if start.elapsed() > timeout {
-            return Err(CoreError::Network(format!(
+            return Err(Error::Network(format!(
                 "Azure Document Intelligence operation polling timed out after {} seconds",
                 timeout.as_secs()
             )));
@@ -373,21 +372,21 @@ pub(super) async fn poll_document_intelligence(
         let response = request_builder
             .send()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         let retry_after = retry_after_secs(&response);
         let status = response.status();
         let text = response
             .text()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
+            .map_err(|err| Error::Network(err.to_string()))?;
         if !status.is_success() {
-            return Err(CoreError::Http {
+            return Err(Error::Http {
                 status: status.as_u16(),
                 body: truncate_error_body(&text),
             });
         }
         let response_json: Value = serde_json::from_str(&text).map_err(|err| {
-            CoreError::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
+            Error::InvalidResponse(format!("invalid Azure DI poll response JSON: {err}"))
         })?;
         if operation_status(&response_json)? == "succeeded" {
             return Ok(response_json);
@@ -426,7 +425,7 @@ mod tests {
 
         assert!(matches!(
             error,
-            CoreError::InvalidRequest(message)
+            Error::InvalidRequest(message)
                 if message.contains("SSRF protection")
         ));
     }
diff --git a/litellm-rust/crates/ai-gateway/src/ocr/handler.rs b/litellm-rust/crates/ai-gateway/src/ocr/handler.rs
index 1de34eb400e..815bc84363a 100644
--- a/litellm-rust/crates/ai-gateway/src/ocr/handler.rs
+++ b/litellm-rust/crates/ai-gateway/src/ocr/handler.rs
@@ -1,5 +1,4 @@
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrResponseHandling;
 use serde_json::Value;
 
@@ -7,7 +6,7 @@ use super::common_utils::{poll_document_intelligence, truncate_error_body};
 use super::types::ProviderOcrRequest;
 use crate::client::http_client;
 
-pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> CoreResult<Value> {
+pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Result<Value, Error> {
     let mut request_builder = http_client().post(&request.url).json(&request.body);
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -19,7 +18,7 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     let status = response.status();
     if request.config.response_handling() == OcrResponseHandling::AzureDocumentIntelligencePoll
@@ -31,7 +30,7 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
             .and_then(|value| value.to_str().ok())
             .map(str::to_string)
             .ok_or_else(|| {
-                CoreError::InvalidResponse(
+                Error::InvalidResponse(
                     "Azure Document Intelligence returned 202 but no Operation-Location header found"
                         .to_string(),
                 )
@@ -52,17 +51,17 @@ pub(crate) async fn execute_ocr_provider_call(request: ProviderOcrRequest) -> Co
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
     let response_json: Value = serde_json::from_str(&text)
-        .map_err(|err| CoreError::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
+        .map_err(|err| Error::InvalidResponse(format!("invalid OCR response JSON: {err}")))?;
 
     Ok(request
         .config
diff --git a/litellm-rust/crates/ai-gateway/src/ocr/hooks.rs b/litellm-rust/crates/ai-gateway/src/ocr/hooks.rs
index 95df566dc53..401e26d3b29 100644
--- a/litellm-rust/crates/ai-gateway/src/ocr/hooks.rs
+++ b/litellm-rust/crates/ai-gateway/src/ocr/hooks.rs
@@ -1,11 +1,9 @@
-use std::future::Future;
-use std::pin::Pin;
-
-use litellm_core::CoreResult;
 use litellm_core::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrAuthStrategy;
 use serde_json::{Map, Value, json};
+use std::future::Future;
+use std::pin::Pin;
 
 use super::common_utils::{
     convert_document_url_to_data_uri, has_header, ocr_provider_config, string_headers,
@@ -27,7 +25,7 @@ pub(crate) struct OcrLifecycleHooks {
     request_metadata: RequestMetadata,
 }
 
-type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type OcrFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 type OcrLogFuture<'a> = Pin<Box<dyn Future<Output = ()> + Send + 'a>>;
 
 impl OcrLifecycleHooks {
@@ -46,7 +44,7 @@ impl OcrLifecycleHooks {
     async fn run_pre_call_guardrails(
         &self,
         request: PreparedOcrRequest,
-    ) -> CoreResult<PreparedOcrRequest> {
+    ) -> Result<PreparedOcrRequest, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(request);
         }
@@ -74,9 +72,9 @@ impl OcrLifecycleHooks {
     async fn prepare_provider_request(
         &self,
         request: PreparedOcrRequest,
-    ) -> CoreResult<ProviderOcrRequest> {
+    ) -> Result<ProviderOcrRequest, Error> {
         let config = ocr_provider_config(&request.custom_llm_provider, &request.model)
-            .ok_or_else(|| CoreError::InvalidProvider(request.custom_llm_provider.clone()))?;
+            .ok_or_else(|| Error::InvalidProvider(request.custom_llm_provider.clone()))?;
         let env_lookup = |key: &str| std::env::var(key).ok();
         let headers = string_headers(request.extra_headers)?;
         let auth_strategy = config.auth_strategy();
@@ -120,7 +118,7 @@ impl OcrLifecycleHooks {
         custom_llm_provider: &str,
         url: &str,
         body: Value,
-    ) -> CoreResult<Value> {
+    ) -> Result<Value, Error> {
         if self.guardrail_runner.is_empty() {
             return Ok(body);
         }
@@ -217,7 +215,7 @@ impl CallLifecycleHooks<PreparedOcrRequest, ProviderOcrRequest, Value> for OcrLi
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -278,19 +276,19 @@ fn guardrail_context(metadata: &RequestMetadata) -> GuardrailContext {
 
 fn parse_ocr_pre_call_guardrail_request(
     request: GuardrailRequest,
-) -> CoreResult<(Value, Map<String, Value>)> {
+) -> Result<(Value, Map<String, Value>), Error> {
     let Value::Object(mut data) = request.data else {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "OCR pre_call guardrail must return an object".to_string(),
         ));
     };
     let document = data.remove("document").ok_or_else(|| {
-        CoreError::InvalidRequest("OCR pre_call guardrail removed document".to_string())
+        Error::InvalidRequest("OCR pre_call guardrail removed document".to_string())
     })?;
     let optional_params = match data.remove("optional_params") {
         Some(Value::Object(params)) => params,
         Some(_) => {
-            return Err(CoreError::InvalidRequest(
+            return Err(Error::InvalidRequest(
                 "OCR pre_call guardrail optional_params must be an object".to_string(),
             ));
         }
@@ -299,33 +297,32 @@ fn parse_ocr_pre_call_guardrail_request(
     Ok((document, optional_params))
 }
 
-fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> CoreResult<Value> {
+fn parse_ocr_during_call_guardrail_request(request: GuardrailRequest) -> Result<Value, Error> {
     let Value::Object(mut data) = request.data else {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "OCR during_call guardrail must return an object".to_string(),
         ));
     };
-    data.remove("body").ok_or_else(|| {
-        CoreError::InvalidRequest("OCR during_call guardrail removed body".to_string())
-    })
+    data.remove("body")
+        .ok_or_else(|| Error::InvalidRequest("OCR during_call guardrail removed body".to_string()))
 }
 
-fn guardrail_error_to_core_error(error: GuardrailError) -> CoreError {
-    CoreError::InvalidRequest(format!("{}: {}", error.kind, error.message))
+fn guardrail_error_to_core_error(error: GuardrailError) -> Error {
+    Error::InvalidRequest(format!("{}: {}", error.kind, error.message))
 }
 
-fn core_error_kind(error: &CoreError) -> &'static str {
+fn core_error_kind(error: &Error) -> &'static str {
     match error {
-        CoreError::Auth(_) => "AuthError",
-        CoreError::InvalidProvider(_) => "InvalidProvider",
-        CoreError::InvalidRequest(_) => "InvalidRequest",
-        CoreError::InvalidType { .. } => "InvalidType",
-        CoreError::MissingField(_) => "MissingField",
-        CoreError::Http { .. } => "HttpError",
-        CoreError::InvalidResponse(_) => "InvalidResponse",
-        CoreError::Network(_) => "NetworkError",
-        CoreError::Connect(_) => "ConnectError",
-        CoreError::Routing(_) => "RoutingError",
-        CoreError::Unsupported(_) => "UnsupportedRequest",
+        Error::Auth(_) => "AuthError",
+        Error::InvalidProvider(_) => "InvalidProvider",
+        Error::InvalidRequest(_) => "InvalidRequest",
+        Error::InvalidType { .. } => "InvalidType",
+        Error::MissingField(_) => "MissingField",
+        Error::Http { .. } => "HttpError",
+        Error::InvalidResponse(_) => "InvalidResponse",
+        Error::Network(_) => "NetworkError",
+        Error::Connect(_) => "ConnectError",
+        Error::Routing(_) => "RoutingError",
+        Error::Unsupported(_) => "UnsupportedRequest",
     }
 }
diff --git a/litellm-rust/crates/ai-gateway/src/ocr/mod.rs b/litellm-rust/crates/ai-gateway/src/ocr/mod.rs
index c4c13e2300c..b59ab626fd3 100644
--- a/litellm-rust/crates/ai-gateway/src/ocr/mod.rs
+++ b/litellm-rust/crates/ai-gateway/src/ocr/mod.rs
@@ -1,4 +1,4 @@
-use litellm_core::CoreResult;
+use litellm_core::Error;
 use litellm_core::call_lifecycle::CallLifecycle;
 use serde_json::Value;
 
@@ -13,7 +13,7 @@ pub use types::OcrRequest;
 use handler::execute_ocr_provider_call;
 use prepare::{PreparedOcrCall, prepare_ocr_call};
 
-pub async fn ocr(request: OcrRequest<'_>) -> CoreResult<Value> {
+pub async fn ocr(request: OcrRequest<'_>) -> Result<Value, Error> {
     let PreparedOcrCall { request, hooks } = prepare_ocr_call(request);
     CallLifecycle::default()
         .run_request(request, &hooks, execute_ocr_provider_call)
diff --git a/litellm-rust/crates/ai-gateway/src/ocr/tests.rs b/litellm-rust/crates/ai-gateway/src/ocr/tests.rs
index bb2a6b06501..8c3f0425149 100644
--- a/litellm-rust/crates/ai-gateway/src/ocr/tests.rs
+++ b/litellm-rust/crates/ai-gateway/src/ocr/tests.rs
@@ -1,7 +1,7 @@
 use std::sync::{Arc, Mutex};
 use std::time::Duration;
 
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::ocr::transformation::OcrResponseHandling;
 use serde_json::{Map, Value, json};
 use tokio::io::{AsyncReadExt, AsyncWriteExt};
@@ -395,7 +395,7 @@ async fn ocr_lifecycle_runs_failure_hook_on_provider_error() {
     .await
     .expect_err("provider error propagates");
 
-    assert!(matches!(err, CoreError::Http { status: 500, .. }));
+    assert!(matches!(err, Error::Http { status: 500, .. }));
     server.await.expect("server task completes");
     assert_eq!(
         logger.events(),
@@ -439,7 +439,7 @@ async fn ocr_lifecycle_pre_call_block_skips_provider_socket() {
     .await
     .expect_err("guardrail blocks request");
 
-    assert!(matches!(err, CoreError::InvalidRequest(_)));
+    assert!(matches!(err, Error::InvalidRequest(_)));
     assert_eq!(guardrail.events(), vec!["async_pre_call_hook"]);
     assert_eq!(
         logger.events(),
@@ -607,7 +607,7 @@ fn string_headers_rejects_non_string_values() {
     let err = string_headers(Some(headers)).expect_err("non-string header rejected");
     assert_eq!(
         err,
-        CoreError::InvalidRequest(
+        Error::InvalidRequest(
             "OCR extra_headers.x-retry-count must be a string, got number".to_string()
         )
     );
diff --git a/litellm-rust/crates/ai-gateway/src/python/config.rs b/litellm-rust/crates/ai-gateway/src/python/config.rs
index c028d3d6b51..d5a4dd69c8d 100644
--- a/litellm-rust/crates/ai-gateway/src/python/config.rs
+++ b/litellm-rust/crates/ai-gateway/src/python/config.rs
@@ -6,33 +6,31 @@
 //! (and recorded in [`crate::gil`]); the realtime hot path never touches Python.
 //!
 //! Compiled only under the `python-config` feature.
-
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::router::{Deployment, Router};
 use pyo3::prelude::*;
 
 use crate::gil;
 
 /// Load the router's `model_list` from `config_path` via the Python reader.
-pub fn load_router_from_config(config_path: &str) -> CoreResult<Router> {
+pub fn load_router_from_config(config_path: &str) -> Result<Router, Error> {
     gil::record_acquisition();
     Python::attach(|py| {
         let model_list = py
             .import("litellm.proxy.read_model_list")
             .and_then(|module| module.getattr("read_model_list"))
             .and_then(|reader| reader.call1((config_path,)))
-            .map_err(|err| CoreError::Routing(format!("read_model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("read_model_list failed: {err}")))?;
 
         let model_list_json: String = py
             .import("json")
             .and_then(|json| json.getattr("dumps"))
             .and_then(|dumps| dumps.call1((model_list,)))
             .and_then(|encoded| encoded.extract())
-            .map_err(|err| CoreError::Routing(format!("serializing model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("serializing model_list failed: {err}")))?;
 
         let deployments: Vec<Deployment> = serde_json::from_str(&model_list_json)
-            .map_err(|err| CoreError::Routing(format!("parsing model_list failed: {err}")))?;
+            .map_err(|err| Error::Routing(format!("parsing model_list failed: {err}")))?;
 
         Ok(Router::new(deployments))
     })
diff --git a/litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs b/litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs
index 7e38d10c6ff..e9f8c477f36 100644
--- a/litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs
+++ b/litellm-rust/crates/ai-gateway/src/routes/messages/mod.rs
@@ -9,7 +9,7 @@ use axum::http::StatusCode;
 use axum::http::header::{CACHE_CONTROL, CONTENT_TYPE, HeaderMap, HeaderValue};
 use axum::response::{IntoResponse, Response};
 use axum::routing::post;
-use litellm_core::CoreError;
+use litellm_core::Error;
 use serde_json::{Map, Value};
 
 use crate::auth::RequireMasterKey;
@@ -46,7 +46,7 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
     let mut response = Response::builder()
         .status(
             StatusCode::from_u16(upstream.status().as_u16()).map_err(|error| {
-                MessagesRouteError(CoreError::InvalidResponse(format!(
+                MessagesRouteError(Error::InvalidResponse(format!(
                     "invalid upstream response status: {error}"
                 )))
             })?,
@@ -58,13 +58,13 @@ fn stream_response(upstream: reqwest::Response) -> Result<Response, MessagesRout
     response
         .body(Body::from_stream(upstream.bytes_stream()))
         .map_err(|error| {
-            MessagesRouteError(CoreError::InvalidResponse(format!(
+            MessagesRouteError(Error::InvalidResponse(format!(
                 "failed to build streaming response: {error}"
             )))
         })
 }
 
-fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, CoreError> {
+fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>, Error> {
     let forwarded = headers
         .iter()
         .filter(|(name, _)| {
@@ -74,19 +74,19 @@ fn forwarded_headers(headers: &HeaderMap) -> Result<Option<Map<String, Value>>,
         })
         .map(|(name, value)| {
             let value = value.to_str().map_err(|_| {
-                CoreError::InvalidRequest(format!("invalid value for header {}", name.as_str()))
+                Error::InvalidRequest(format!("invalid value for header {}", name.as_str()))
             })?;
             Ok((name.to_string(), Value::String(value.to_string())))
         })
-        .collect::<Result<Map<_, _>, CoreError>>()?;
+        .collect::<Result<Map<_, _>, Error>>()?;
     Ok((!forwarded.is_empty()).then_some(forwarded))
 }
 
 #[derive(Debug)]
-struct MessagesRouteError(CoreError);
+struct MessagesRouteError(Error);
 
-impl From<CoreError> for MessagesRouteError {
-    fn from(error: CoreError) -> Self {
+impl From<Error> for MessagesRouteError {
+    fn from(error: Error) -> Self {
         Self(error)
     }
 }
@@ -94,28 +94,28 @@ impl From<CoreError> for MessagesRouteError {
 impl IntoResponse for MessagesRouteError {
     fn into_response(self) -> Response {
         let (status, message) = match self.0 {
-            CoreError::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
-            CoreError::InvalidProvider(_) | CoreError::Routing(_) => (
+            Error::InvalidRequest(message) => (StatusCode::BAD_REQUEST, message),
+            Error::InvalidProvider(_) | Error::Routing(_) => (
                 StatusCode::NOT_FOUND,
                 "no messages deployment is configured for this model".to_string(),
             ),
-            CoreError::Auth(_) => (
+            Error::Auth(_) => (
                 StatusCode::BAD_GATEWAY,
                 "messages provider authentication failed".to_string(),
             ),
-            CoreError::Http { .. }
-            | CoreError::Network(_)
-            | CoreError::Connect(_)
-            | CoreError::InvalidResponse(_)
-            | CoreError::InvalidType { .. }
-            | CoreError::MissingField(_) => (
+            Error::Http { .. }
+            | Error::Network(_)
+            | Error::Connect(_)
+            | Error::InvalidResponse(_)
+            | Error::InvalidType { .. }
+            | Error::MissingField(_) => (
                 StatusCode::BAD_GATEWAY,
                 "messages provider request failed".to_string(),
             ),
             // The gateway has no Python implementation to decline to, so a
             // request the core cannot serve is reported to the caller. The
             // reason is a fixed internal string, never provider content.
-            CoreError::Unsupported(reason) => (
+            Error::Unsupported(reason) => (
                 StatusCode::BAD_REQUEST,
                 format!("messages request is not supported: {reason}"),
             ),
diff --git a/litellm-rust/crates/ai-gateway/src/routes/messages/service.rs b/litellm-rust/crates/ai-gateway/src/routes/messages/service.rs
index 5f4c5fe8de4..4fd29db05d6 100644
--- a/litellm-rust/crates/ai-gateway/src/routes/messages/service.rs
+++ b/litellm-rust/crates/ai-gateway/src/routes/messages/service.rs
@@ -1,10 +1,10 @@
 use std::sync::Arc;
 
+use litellm_core::Error;
 use litellm_core::constants::ANTHROPIC_MESSAGES_PROVIDER;
 use litellm_core::messages::types::MessagesRequest;
 use litellm_core::messages::{messages, messages_stream};
 use litellm_core::router::Router;
-use litellm_core::{CoreError, CoreResult};
 use serde_json::{Map, Value};
 
 pub(crate) enum MessagesResponse {
@@ -16,16 +16,16 @@ pub async fn run(
     router: &Arc<Router>,
     body: Value,
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<MessagesResponse> {
+) -> Result<MessagesResponse, Error> {
     let model = body
         .get("model")
         .and_then(Value::as_str)
         .map(str::trim)
         .filter(|model| !model.is_empty())
-        .ok_or_else(|| CoreError::InvalidRequest("messages body requires a model".to_string()))?;
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+        .ok_or_else(|| Error::InvalidRequest("messages body requires a model".to_string()))?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let provider_model = deployment.litellm_params.model.as_str();
     let upstream_model = provider_model
         .split_once('/')
@@ -37,7 +37,7 @@ pub async fn run(
     };
     let mut body = body;
     body.as_object_mut()
-        .ok_or_else(|| CoreError::InvalidRequest("messages body must be an object".to_string()))?
+        .ok_or_else(|| Error::InvalidRequest("messages body must be an object".to_string()))?
         .insert(
             "model".to_string(),
             Value::String(upstream_model.to_string()),
@@ -60,6 +60,6 @@ pub async fn run(
     serde_json::to_value(response)
         .map(MessagesResponse::Json)
         .map_err(|err| {
-            CoreError::InvalidResponse(format!("failed to serialize messages response: {err}"))
+            Error::InvalidResponse(format!("failed to serialize messages response: {err}"))
         })
 }
diff --git a/litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs b/litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs
index 4ae8cfe7379..b8ee77c4269 100644
--- a/litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs
+++ b/litellm-rust/crates/ai-gateway/src/routes/realtime/service.rs
@@ -11,8 +11,7 @@ use std::time::Duration;
 
 use crate::io::realtime_pool::{RealtimePool, upstream_key};
 use futures_util::{Sink, Stream};
-use litellm_core::CoreResult;
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::realtime::types::RealtimeEvent;
 use litellm_core::router::Router;
 
@@ -29,15 +28,15 @@ pub async fn run<In, Out>(
     observe: impl FnMut(&RealtimeEvent) + Send,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = RealtimeEvent> + Unpin + Send,
     Out: Sink<RealtimeEvent> + Unpin + Send,
     <Out as Sink<RealtimeEvent>>::Error: std::fmt::Display,
 {
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let params = &deployment.litellm_params;
     // Strip a leading `openai/` so the OpenAI-only realtime fn gets the bare model.
     let provider_model = params
diff --git a/litellm-rust/crates/ai-gateway/src/routes/responses/service.rs b/litellm-rust/crates/ai-gateway/src/routes/responses/service.rs
index 165c95695d3..e8f840c0c8e 100644
--- a/litellm-rust/crates/ai-gateway/src/routes/responses/service.rs
+++ b/litellm-rust/crates/ai-gateway/src/routes/responses/service.rs
@@ -2,13 +2,13 @@ use std::sync::Arc;
 use std::time::Duration;
 
 use futures_util::{Sink, Stream};
+use litellm_core::Error;
 use litellm_core::call_lifecycle::{CallLifecycle, CallLifecycleContext};
 use litellm_core::responses::instrumentation::{
     ResponsesWsCallbackPayload, ResponsesWsInstrumentation, ResponsesWsLogOutcome,
     ResponsesWsMetadata,
 };
 use litellm_core::responses::types::ResponsesWsEvent;
-use litellm_core::{CoreError, CoreResult};
 
 use crate::integrations::custom_logger::{
     CallbackTiming, CallbackValue, CustomLogger, CustomLoggerRunner, LoggingError, ModelCallDetails,
@@ -26,22 +26,22 @@ pub async fn run<In, Out>(
     metadata: RequestMetadata,
     client_in: In,
     client_out: Out,
-) -> CoreResult<()>
+) -> Result<(), Error>
 where
     In: Stream<Item = ResponsesWsEvent> + Unpin + Send,
     Out: Sink<ResponsesWsEvent> + Unpin + Send,
     Out::Error: std::fmt::Display,
 {
-    let deployment = router.get_available_deployment(model).ok_or_else(|| {
-        CoreError::Routing(format!("no deployment available for model '{model}'"))
-    })?;
+    let deployment = router
+        .get_available_deployment(model)
+        .ok_or_else(|| Error::Routing(format!("no deployment available for model '{model}'")))?;
     let params = &deployment.litellm_params;
     let provider_model = params
         .model
         .strip_prefix("openai/")
         .unwrap_or(&params.model);
     if params.model.contains('/') && !params.model.starts_with("openai/") {
-        return Err(CoreError::InvalidProvider(
+        return Err(Error::InvalidProvider(
             "Responses WebSocket route supports OpenAI deployments only".to_string(),
         ));
     }
diff --git a/litellm-rust/crates/core/src/audio_transcription/transformation.rs b/litellm-rust/crates/core/src/audio_transcription/transformation.rs
index eab34c13843..16a28fbcac0 100644
--- a/litellm-rust/crates/core/src/audio_transcription/transformation.rs
+++ b/litellm-rust/crates/core/src/audio_transcription/transformation.rs
@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::CoreResult;
-
 use super::types::{AudioTranscriptionRequestData, AudioTranscriptionResponseData};
 
 #[derive(Clone, Debug, PartialEq, Eq)]
@@ -32,13 +31,13 @@ pub trait AudioTranscriptionProviderConfig: Sync {
         model: &str,
         audio: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<AudioTranscriptionRequestData>;
+    ) -> Result<AudioTranscriptionRequestData, Error>;
 
     fn transform_transcription_response(
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<AudioTranscriptionResponseData>;
+    ) -> Result<AudioTranscriptionResponseData, Error>;
 
     fn complete_url(
         &self,
@@ -46,12 +45,12 @@ pub trait AudioTranscriptionProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(
         &self,
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<AudioTranscriptionAuth>;
+    ) -> Result<AudioTranscriptionAuth, Error>;
 }
diff --git a/litellm-rust/crates/core/src/call_lifecycle/mod.rs b/litellm-rust/crates/core/src/call_lifecycle/mod.rs
index d9b68a1b726..637c156e192 100644
--- a/litellm-rust/crates/core/src/call_lifecycle/mod.rs
+++ b/litellm-rust/crates/core/src/call_lifecycle/mod.rs
@@ -1,7 +1,7 @@
 use std::future::Future;
 use std::time::{Instant, SystemTime, UNIX_EPOCH};
 
-use crate::{CoreError, CoreResult};
+use crate::Error;
 
 pub mod types;
 
@@ -11,14 +11,14 @@ pub use types::{
 };
 
 pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
-    type PreCallFuture<'a>: Future<Output = CoreResult<InitialReq>> + Send + 'a
+    type PreCallFuture<'a>: Future<Output = Result<InitialReq, Error>> + Send + 'a
     where
         Self: 'a,
         InitialReq: 'a,
         ProviderReq: 'a,
         Resp: 'a;
 
-    type DuringCallFuture<'a>: Future<Output = CoreResult<ProviderReq>> + Send + 'a
+    type DuringCallFuture<'a>: Future<Output = Result<ProviderReq, Error>> + Send + 'a
     where
         Self: 'a,
         InitialReq: 'a,
@@ -56,7 +56,7 @@ pub trait CallLifecycleHooks<InitialReq, ProviderReq, Resp>: Send + Sync {
     fn async_log_failure_event<'a>(
         &'a self,
         context: &'a CallLifecycleContext,
-        error: &'a CoreError,
+        error: &'a Error,
         timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a>;
 }
@@ -86,12 +86,12 @@ impl<'a> CallLifecycle<'a> {
         request: InitialReq,
         hooks: &Hooks,
         provider_call: ProviderCall,
-    ) -> CoreResult<Resp>
+    ) -> Result<Resp, Error>
     where
         InitialReq: CallLifecycleRequest,
         Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
         ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
-        ProviderFuture: Future<Output = CoreResult<Resp>>,
+        ProviderFuture: Future<Output = Result<Resp, Error>>,
     {
         let context = request.lifecycle_context();
         self.run(context, request, hooks, provider_call).await
@@ -103,11 +103,11 @@ impl<'a> CallLifecycle<'a> {
         request: InitialReq,
         hooks: &Hooks,
         provider_call: ProviderCall,
-    ) -> CoreResult<Resp>
+    ) -> Result<Resp, Error>
     where
         Hooks: CallLifecycleHooks<InitialReq, ProviderReq, Resp>,
         ProviderCall: FnOnce(ProviderReq) -> ProviderFuture,
-        ProviderFuture: Future<Output = CoreResult<Resp>>,
+        ProviderFuture: Future<Output = Result<Resp, Error>>,
     {
         let call_start = epoch_seconds();
         let mut phases = Vec::new();
@@ -166,7 +166,7 @@ impl<'a> CallLifecycle<'a> {
         &self,
         context: &CallLifecycleContext,
         hooks: &Hooks,
-        error: &CoreError,
+        error: &Error,
         call_start: f64,
         phases: &mut Vec<CallLifecyclePhaseTiming>,
     ) where
@@ -251,8 +251,8 @@ mod tests {
     }
 
     impl CallLifecycleHooks<String, String, String> for RecordingHooks {
-        type PreCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
-        type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
+        type PreCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
+        type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
         type SuccessFuture<'a> = BoxFuture<'a, ()>;
         type FailureFuture<'a> = BoxFuture<'a, ()>;
 
@@ -294,7 +294,7 @@ mod tests {
         fn async_log_failure_event<'a>(
             &'a self,
             _context: &'a CallLifecycleContext,
-            _error: &'a CoreError,
+            _error: &'a Error,
             _timing: &'a CallLifecycleTiming,
         ) -> Self::FailureFuture<'a> {
             Box::pin(async move {
@@ -304,8 +304,8 @@ mod tests {
     }
 
     impl CallLifecycleHooks<RecordingRequest, String, String> for RecordingHooks {
-        type PreCallFuture<'a> = BoxFuture<'a, CoreResult<RecordingRequest>>;
-        type DuringCallFuture<'a> = BoxFuture<'a, CoreResult<String>>;
+        type PreCallFuture<'a> = BoxFuture<'a, Result<RecordingRequest, Error>>;
+        type DuringCallFuture<'a> = BoxFuture<'a, Result<String, Error>>;
         type SuccessFuture<'a> = BoxFuture<'a, ()>;
         type FailureFuture<'a> = BoxFuture<'a, ()>;
 
@@ -345,7 +345,7 @@ mod tests {
         fn async_log_failure_event<'a>(
             &'a self,
             _context: &'a CallLifecycleContext,
-            _error: &'a CoreError,
+            _error: &'a Error,
             _timing: &'a CallLifecycleTiming,
         ) -> Self::FailureFuture<'a> {
             Box::pin(async move {
@@ -383,13 +383,13 @@ mod tests {
                 "request".to_string(),
                 &hooks,
                 |_request| async move {
-                    Err::<String, CoreError>(CoreError::Network("provider down".to_string()))
+                    Err::<String, Error>(Error::Network("provider down".to_string()))
                 },
             )
             .await
             .expect_err("call fails");
 
-        assert_eq!(error, CoreError::Network("provider down".to_string()));
+        assert_eq!(error, Error::Network("provider down".to_string()));
         assert_eq!(hooks.events(), vec!["pre_call", "during_call", "failure"]);
     }
 
diff --git a/litellm-rust/crates/core/src/chat_completions/common_utils.rs b/litellm-rust/crates/core/src/chat_completions/common_utils.rs
index 36eaf242a5a..ca51471eb7c 100644
--- a/litellm-rust/crates/core/src/chat_completions/common_utils.rs
+++ b/litellm-rust/crates/core/src/chat_completions/common_utils.rs
@@ -1,8 +1,7 @@
-use serde_json::{Map, Value};
-
-use crate::error::CoreResult;
+use crate::Error;
 use crate::http_utils::string_headers as shared_string_headers;
 use crate::providers::anthropic::chat_completions::transformation::ANTHROPIC_CHAT_COMPLETIONS_CONFIG;
+use serde_json::{Map, Value};
 
 use super::transformation::ChatCompletionsProviderConfig;
 
@@ -23,6 +22,6 @@ pub(super) fn chat_completions_provider_config(
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     shared_string_headers(HEADER_CONTEXT, extra_headers)
 }
diff --git a/litellm-rust/crates/core/src/chat_completions/handler.rs b/litellm-rust/crates/core/src/chat_completions/handler.rs
index afc4529fd26..7e2731442cc 100644
--- a/litellm-rust/crates/core/src/chat_completions/handler.rs
+++ b/litellm-rust/crates/core/src/chat_completions/handler.rs
@@ -1,6 +1,6 @@
 use serde_json::Value;
 
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::http_utils::truncate_error_body;
 
 use super::client::http_client;
@@ -11,9 +11,9 @@ use super::types::{
 
 pub(super) async fn execute_chat_completions_provider_call(
     request: ProviderChatCompletionsRequest,
-) -> CoreResult<ChatCompletionsResponse> {
+) -> Result<ChatCompletionsResponse, Error> {
     let body = serde_json::to_vec(&request.body).map_err(|err| {
-        CoreError::InvalidRequest(format!(
+        Error::InvalidRequest(format!(
             "failed to serialize chat completions request: {err}"
         ))
     })?;
@@ -32,9 +32,9 @@ pub(super) async fn execute_chat_completions_provider_call(
         // so the host can still serve it. Everything else here, a timeout
         // above all, may have reached the provider and been answered.
         if err.is_connect() || err.is_builder() {
-            CoreError::Connect(err.to_string())
+            Error::Connect(err.to_string())
         } else {
-            CoreError::Network(err.to_string())
+            Error::Network(err.to_string())
         }
     })?;
 
@@ -42,17 +42,17 @@ pub(super) async fn execute_chat_completions_provider_call(
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
     let body: Value = serde_json::from_str(&text).map_err(|err| {
-        CoreError::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
+        Error::InvalidResponse(format!("invalid chat completions response JSON: {err}"))
     })?;
     request
         .config
@@ -69,10 +69,10 @@ pub(super) async fn execute_chat_completions_provider_call(
 /// second kind has already been billed, and a host that keeps a reference
 /// implementation must not retry those, so collapse them to one variant that
 /// can only mean the provider was already called.
-pub(super) fn as_response_error(err: CoreError) -> CoreError {
+pub(super) fn as_response_error(err: Error) -> Error {
     match err {
-        already @ (CoreError::InvalidResponse(_) | CoreError::Http { .. }) => already,
-        other => CoreError::InvalidResponse(other.to_string()),
+        already @ (Error::InvalidResponse(_) | Error::Http { .. }) => already,
+        other => Error::InvalidResponse(other.to_string()),
     }
 }
 
@@ -80,7 +80,7 @@ pub(super) fn as_response_error(err: CoreError) -> CoreError {
 pub(super) async fn signed_headers(
     request: &ProviderChatCompletionsRequest,
     body: &[u8],
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     use std::collections::BTreeMap;
     use std::time::SystemTime;
 
@@ -101,7 +101,7 @@ pub(super) async fn signed_headers(
         .iter()
         .any(|(name, _)| is_sigv4_computed_header(name))
     {
-        return Err(CoreError::Unsupported(
+        return Err(Error::Unsupported(
             "request forwards a header AWS SigV4 computes",
         ));
     }
@@ -137,9 +137,9 @@ pub(super) async fn signed_headers(
 pub(super) async fn signed_headers(
     request: &ProviderChatCompletionsRequest,
     _body: &[u8],
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     match &request.auth {
-        ChatCompletionsAuth::AwsSigV4 { .. } => Err(CoreError::Unsupported(
+        ChatCompletionsAuth::AwsSigV4 { .. } => Err(Error::Unsupported(
             "AWS SigV4 requires the bedrock-auth feature",
         )),
         _ => Ok(request.upstream_headers.clone()),
diff --git a/litellm-rust/crates/core/src/chat_completions/mod.rs b/litellm-rust/crates/core/src/chat_completions/mod.rs
index f30ac1a24bf..0d009d36d16 100644
--- a/litellm-rust/crates/core/src/chat_completions/mod.rs
+++ b/litellm-rust/crates/core/src/chat_completions/mod.rs
@@ -6,6 +6,7 @@
 //! credentials, and it resolves the provider, translates the conversation,
 //! calls the provider, and returns a typed OpenAI-shaped response.
 
+use crate::Error;
 mod client;
 mod common_utils;
 pub mod conversation;
@@ -17,15 +18,13 @@ pub mod types;
 
 use serde_json::{Map, Value};
 
-use crate::error::CoreResult;
-
 use handler::execute_chat_completions_provider_call;
 use prepare::{parse_messages, prepare_chat_completions_call, resolve_provider_config};
 use types::{ChatCompletionsRequest, ChatCompletionsResponse};
 
 pub async fn chat_completions(
     request: ChatCompletionsRequest<'_>,
-) -> CoreResult<ChatCompletionsResponse> {
+) -> Result<ChatCompletionsResponse, Error> {
     execute_chat_completions_provider_call(prepare_chat_completions_call(request)?).await
 }
 
diff --git a/litellm-rust/crates/core/src/chat_completions/prepare.rs b/litellm-rust/crates/core/src/chat_completions/prepare.rs
index 1e1c8d1bafd..142b2f2aaed 100644
--- a/litellm-rust/crates/core/src/chat_completions/prepare.rs
+++ b/litellm-rust/crates/core/src/chat_completions/prepare.rs
@@ -1,6 +1,6 @@
 use serde_json::Value;
 
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::http_utils::has_header;
 use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
 
@@ -11,7 +11,7 @@ use super::types::{ChatCompletionsRequest, ChatMessage, ProviderChatCompletionsR
 pub(super) fn resolve_provider_config<'a>(
     model: &'a str,
     custom_llm_provider: Option<&'a str>,
-) -> CoreResult<(String, &'static dyn ChatCompletionsProviderConfig)> {
+) -> Result<(String, &'static dyn ChatCompletionsProviderConfig), Error> {
     let provider_info = get_custom_llm_provider(model, custom_llm_provider)
         .or_else(|| {
             custom_llm_provider.map(|provider| CustomLlmProvider {
@@ -20,35 +20,34 @@ pub(super) fn resolve_provider_config<'a>(
             })
         })
         .ok_or_else(|| {
-            CoreError::InvalidProvider(
+            Error::InvalidProvider(
                 "unable to resolve custom_llm_provider for chat completions request".to_string(),
             )
         })?;
     let config = chat_completions_provider_config(provider_info.custom_llm_provider)
-        .ok_or_else(|| CoreError::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
+        .ok_or_else(|| Error::InvalidProvider(provider_info.custom_llm_provider.to_string()))?;
     Ok((provider_info.model.to_string(), config))
 }
 
-pub(super) fn parse_messages(messages: Value) -> CoreResult<Vec<ChatMessage>> {
-    serde_json::from_value(messages).map_err(|err| {
-        CoreError::InvalidRequest(format!("invalid chat completions messages: {err}"))
-    })
+pub(super) fn parse_messages(messages: Value) -> Result<Vec<ChatMessage>, Error> {
+    serde_json::from_value(messages)
+        .map_err(|err| Error::InvalidRequest(format!("invalid chat completions messages: {err}")))
 }
 
 pub(super) fn prepare_chat_completions_call(
     request: ChatCompletionsRequest<'_>,
-) -> CoreResult<ProviderChatCompletionsRequest> {
+) -> Result<ProviderChatCompletionsRequest, Error> {
     let (model, config) = resolve_provider_config(request.model, request.custom_llm_provider)?;
     let env_lookup = |key: &str| std::env::var(key).ok();
 
     let messages = parse_messages(request.messages)?;
     if messages.is_empty() {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "chat completions requires at least one message".to_string(),
         ));
     }
     if let Some(reason) = config.unsupported_reason(&messages, &request.optional_params) {
-        return Err(CoreError::Unsupported(reason.0));
+        return Err(Error::Unsupported(reason.0));
     }
 
     let mut headers = string_headers(request.extra_headers)?;
diff --git a/litellm-rust/crates/core/src/chat_completions/tests.rs b/litellm-rust/crates/core/src/chat_completions/tests.rs
index e2383723cb0..2858d180e27 100644
--- a/litellm-rust/crates/core/src/chat_completions/tests.rs
+++ b/litellm-rust/crates/core/src/chat_completions/tests.rs
@@ -1,6 +1,6 @@
 use serde_json::{Map, Value, json};
 
-use crate::error::CoreError;
+use crate::error::Error;
 
 use super::prepare::prepare_chat_completions_call;
 use super::transformation::ChatCompletionsAuth;
@@ -29,7 +29,7 @@ fn request<'a>(
 
 /// `ProviderChatCompletionsRequest` deliberately has no `Debug` (its headers
 /// carry resolved credentials), so unwrap the failure case by hand.
-fn decline(request: ChatCompletionsRequest<'_>) -> CoreError {
+fn decline(request: ChatCompletionsRequest<'_>) -> Error {
     match prepare_chat_completions_call(request) {
         Err(error) => error,
         Ok(prepared) => panic!("expected a decline, prepared a call to {}", prepared.url),
@@ -196,7 +196,7 @@ fn declines_an_unsupported_request_before_resolving_credentials() {
     call.api_key = None;
     // No api_key is set and no env is consulted: the gate must run first, so the
     // error is the decline rather than a missing-credential error.
-    assert_eq!(decline(call), CoreError::Unsupported("streaming"));
+    assert_eq!(decline(call), Error::Unsupported("streaming"));
 }
 
 #[test]
@@ -208,7 +208,7 @@ fn rejects_an_unknown_provider() {
             json!([{"role": "user", "content": "hi"}]),
             json!({}),
         )),
-        CoreError::InvalidProvider("openai".to_string())
+        Error::InvalidProvider("openai".to_string())
     );
 }
 
@@ -221,7 +221,7 @@ fn rejects_a_model_with_no_resolvable_provider() {
             json!([{"role": "user", "content": "hi"}]),
             json!({}),
         )),
-        CoreError::InvalidProvider(_)
+        Error::InvalidProvider(_)
     ));
 }
 
@@ -234,7 +234,7 @@ fn rejects_an_empty_or_malformed_message_list() {
             json!([]),
             json!({}),
         )),
-        CoreError::InvalidRequest("chat completions requires at least one message".to_string())
+        Error::InvalidRequest("chat completions requires at least one message".to_string())
     );
     assert!(matches!(
         decline(request(
@@ -243,7 +243,7 @@ fn rejects_an_empty_or_malformed_message_list() {
             json!("not a list"),
             json!({}),
         )),
-        CoreError::InvalidRequest(_)
+        Error::InvalidRequest(_)
     ));
 }
 
@@ -258,7 +258,7 @@ fn rejects_non_string_extra_headers() {
     call.extra_headers = Some(Map::from_iter([("x-trace".to_string(), json!(7))]));
     assert_eq!(
         decline(call),
-        CoreError::InvalidRequest(
+        Error::InvalidRequest(
             "chat completions extra_headers.x-trace must be a string, got number".to_string()
         )
     );
@@ -374,7 +374,7 @@ async fn a_forwarded_header_the_signer_computes_declines_to_python() {
             .await
             .expect_err("{forwarded} should decline instead of being signed");
         assert!(
-            matches!(error, CoreError::Unsupported(_)),
+            matches!(error, Error::Unsupported(_)),
             "{forwarded} declined as {error:?}, which the host would not fall back on"
         );
     }
@@ -727,7 +727,7 @@ mod round_trip {
         .expect_err("response cannot be normalized");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::InvalidResponse(_)),
+            matches!(err, Error::InvalidResponse(_)),
             "expected a post-send error, got {err:?}"
         );
     }
@@ -745,7 +745,7 @@ mod round_trip {
         .expect_err("response cannot be normalized");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::InvalidResponse(_)),
+            matches!(err, Error::InvalidResponse(_)),
             "expected a post-send error, got {err:?}"
         );
     }
@@ -763,7 +763,7 @@ mod round_trip {
         .expect_err("upstream rejects");
         handle.await.expect("server task");
         assert!(
-            matches!(err, CoreError::Http { status: 429, .. }),
+            matches!(err, Error::Http { status: 429, .. }),
             "expected a 429, got {err:?}"
         );
     }
@@ -787,7 +787,7 @@ mod round_trip {
         .await
         .expect_err("nothing is listening");
         assert!(
-            matches!(err, CoreError::Connect(_)),
+            matches!(err, Error::Connect(_)),
             "expected a pre-send connect failure, got {err:?}"
         );
     }
@@ -797,24 +797,24 @@ mod round_trip {
         use crate::chat_completions::handler::as_response_error;
 
         for original in [
-            CoreError::MissingField("usage"),
-            CoreError::Unsupported("non-text response content block"),
-            CoreError::InvalidRequest("whatever".to_string()),
-            CoreError::Auth("whatever".to_string()),
+            Error::MissingField("usage"),
+            Error::Unsupported("non-text response content block"),
+            Error::InvalidRequest("whatever".to_string()),
+            Error::Auth("whatever".to_string()),
         ] {
             let label = format!("{original:?}");
             assert!(
-                matches!(as_response_error(original), CoreError::InvalidResponse(_)),
+                matches!(as_response_error(original), Error::InvalidResponse(_)),
                 "{label} must not stay retryable once the provider has answered"
             );
         }
         // An upstream status is already unambiguous, so it survives intact.
         assert!(matches!(
-            as_response_error(CoreError::Http {
+            as_response_error(Error::Http {
                 status: 500,
                 body: "boom".to_string()
             }),
-            CoreError::Http { status: 500, .. }
+            Error::Http { status: 500, .. }
         ));
     }
 }
diff --git a/litellm-rust/crates/core/src/chat_completions/transformation.rs b/litellm-rust/crates/core/src/chat_completions/transformation.rs
index a30ce9dc77c..a0868209305 100644
--- a/litellm-rust/crates/core/src/chat_completions/transformation.rs
+++ b/litellm-rust/crates/core/src/chat_completions/transformation.rs
@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::error::CoreResult;
-
 use super::types::{
     ChatCompletionsResponse, ChatMessage, ChatMessageContent, ProviderChatRequestData,
     ProviderChatResponseData,
@@ -39,7 +38,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth(
         &self,
@@ -47,7 +46,7 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth>;
+    ) -> Result<ChatCompletionsAuth, Error>;
 
     fn default_headers(&self) -> &'static [(&'static str, &'static str)] {
         &[("content-type", "application/json")]
@@ -91,13 +90,13 @@ pub trait ChatCompletionsProviderConfig: Sync {
         model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData>;
+    ) -> Result<ProviderChatRequestData, Error>;
 
     fn transform_response(
         &self,
         model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse>;
+    ) -> Result<ChatCompletionsResponse, Error>;
 }
 
 pub fn unsupported_param(
diff --git a/litellm-rust/crates/core/src/error.rs b/litellm-rust/crates/core/src/error.rs
index 739532f8cb5..db3fa2ec704 100644
--- a/litellm-rust/crates/core/src/error.rs
+++ b/litellm-rust/crates/core/src/error.rs
@@ -1,9 +1,7 @@
-use thiserror::Error;
+use thiserror::Error as ThisError;
 
-pub type CoreResult<T> = Result<T, CoreError>;
-
-#[derive(Debug, Error, PartialEq, Eq)]
-pub enum CoreError {
+#[derive(Debug, ThisError, PartialEq, Eq)]
+pub enum Error {
     #[error("expected {expected}, got {actual}")]
     InvalidType {
         expected: &'static str,
diff --git a/litellm-rust/crates/core/src/http_utils.rs b/litellm-rust/crates/core/src/http_utils.rs
index c541f50275b..10661fadf96 100644
--- a/litellm-rust/crates/core/src/http_utils.rs
+++ b/litellm-rust/crates/core/src/http_utils.rs
@@ -3,7 +3,7 @@
 use serde_json::{Map, Value};
 
 use crate::constants::UPSTREAM_ERROR_BODY_MAX_CHARS;
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 
 /// Bound an upstream error body before it crosses a host boundary, so provider
 /// bodies stay data-minimized.
@@ -18,7 +18,7 @@ pub fn truncate_error_body(body: &str) -> String {
 pub fn string_headers(
     context: &'static str,
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     extra_headers
         .unwrap_or_default()
         .into_iter()
@@ -27,7 +27,7 @@ pub fn string_headers(
                 .as_str()
                 .map(|value| (key.clone(), value.to_string()))
                 .ok_or_else(|| {
-                    CoreError::InvalidRequest(format!(
+                    Error::InvalidRequest(format!(
                         "{context} extra_headers.{key} must be a string, got {}",
                         json_type_name(&value)
                     ))
@@ -81,7 +81,7 @@ mod tests {
         let err = string_headers("chat completions", Some(headers)).expect_err("non-string value");
         assert_eq!(
             err,
-            CoreError::InvalidRequest(
+            Error::InvalidRequest(
                 "chat completions extra_headers.x-trace must be a string, got number".to_string()
             )
         );
diff --git a/litellm-rust/crates/core/src/lib.rs b/litellm-rust/crates/core/src/lib.rs
index dce4a425ea0..0e18d24e5d8 100644
--- a/litellm-rust/crates/core/src/lib.rs
+++ b/litellm-rust/crates/core/src/lib.rs
@@ -13,4 +13,4 @@ pub mod responses;
 pub mod router;
 pub mod routing_utils;
 
-pub use error::{CoreError, CoreResult};
+pub use error::Error;
diff --git a/litellm-rust/crates/core/src/messages/common_utils.rs b/litellm-rust/crates/core/src/messages/common_utils.rs
index a14dffbc1fe..8dfdb2e361a 100644
--- a/litellm-rust/crates/core/src/messages/common_utils.rs
+++ b/litellm-rust/crates/core/src/messages/common_utils.rs
@@ -1,9 +1,8 @@
-use serde_json::{Map, Value};
-
-use crate::error::CoreResult;
+use crate::Error;
 use crate::http_utils::string_headers as shared_string_headers;
 use crate::providers::anthropic::messages::transformation::ANTHROPIC_MESSAGES_CONFIG;
 use crate::providers::azure_ai::messages::transformation::AZURE_ANTHROPIC_MESSAGES_CONFIG;
+use serde_json::{Map, Value};
 
 use super::transformation::AnthropicMessagesProviderConfig;
 
@@ -23,6 +22,6 @@ pub(super) fn messages_provider_config(
 
 pub(super) fn string_headers(
     extra_headers: Option<Map<String, Value>>,
-) -> CoreResult<Vec<(String, String)>> {
+) -> Result<Vec<(String, String)>, Error> {
     shared_string_headers(HEADER_CONTEXT, extra_headers)
 }
diff --git a/litellm-rust/crates/core/src/messages/handler.rs b/litellm-rust/crates/core/src/messages/handler.rs
index 1c895f66eba..13a65d86131 100644
--- a/litellm-rust/crates/core/src/messages/handler.rs
+++ b/litellm-rust/crates/core/src/messages/handler.rs
@@ -1,5 +1,5 @@
 use crate::constants::ANTHROPIC_MESSAGES_PROVIDER;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 
 use super::client::http_client;
 use super::common_utils::truncate_error_body;
@@ -7,7 +7,7 @@ use super::types::{AnthropicMessagesResponse, ProviderMessagesRequest};
 
 pub(super) async fn execute_messages_provider_call(
     request: ProviderMessagesRequest,
-) -> CoreResult<AnthropicMessagesResponse> {
+) -> Result<AnthropicMessagesResponse, Error> {
     let mut request_builder = http_client().post(&request.url).json(&request.body);
     for (key, value) in &request.upstream_headers {
         request_builder = request_builder.header(key, value);
@@ -19,32 +19,31 @@ pub(super) async fn execute_messages_provider_call(
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     let status = response.status();
     let text = response
         .text()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
 
     if !status.is_success() {
-        return Err(CoreError::Http {
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
     }
 
-    let response = serde_json::from_str(&text).map_err(|err| {
-        CoreError::InvalidResponse(format!("invalid messages response JSON: {err}"))
-    })?;
+    let response = serde_json::from_str(&text)
+        .map_err(|err| Error::InvalidResponse(format!("invalid messages response JSON: {err}")))?;
     request.config.transform_response(&request.model, response)
 }
 
 pub(super) async fn execute_messages_provider_stream(
     request: ProviderMessagesRequest,
-) -> CoreResult<reqwest::Response> {
+) -> Result<reqwest::Response, Error> {
     if request.provider != ANTHROPIC_MESSAGES_PROVIDER {
-        return Err(CoreError::InvalidRequest(
+        return Err(Error::InvalidRequest(
             "streaming messages is not supported for this provider".to_string(),
         ));
     }
@@ -60,14 +59,14 @@ pub(super) async fn execute_messages_provider_stream(
     let response = request_builder
         .send()
         .await
-        .map_err(|err| CoreError::Network(err.to_string()))?;
+        .map_err(|err| Error::Network(err.to_string()))?;
     let status = response.status();
     if !status.is_success() {
         let text = response
             .text()
             .await
-            .map_err(|err| CoreError::Network(err.to_string()))?;
-        return Err(CoreError::Http {
+            .map_err(|err| Error::Network(err.to_string()))?;
+        return Err(Error::Http {
             status: status.as_u16(),
             body: truncate_error_body(&text),
         });
diff --git a/litellm-rust/crates/core/src/messages/mod.rs b/litellm-rust/crates/core/src/messages/mod.rs
index acb36d89daf..ee2877e61fc 100644
--- a/litellm-rust/crates/core/src/messages/mod.rs
+++ b/litellm-rust/crates/core/src/messages/mod.rs
@@ -7,6 +7,7 @@
 //! is the streaming variant; it hands the raw upstream response back so a host
 //! can splice the event stream to its own caller.
 
+use crate::Error;
 mod client;
 mod common_utils;
 mod handler;
@@ -14,17 +15,15 @@ mod prepare;
 pub mod transformation;
 pub mod types;
 
-use crate::error::CoreResult;
-
 use handler::{execute_messages_provider_call, execute_messages_provider_stream};
 use prepare::prepare_messages_call;
 use types::{AnthropicMessagesResponse, MessagesRequest};
 
-pub async fn messages(request: MessagesRequest<'_>) -> CoreResult<AnthropicMessagesResponse> {
+pub async fn messages(request: MessagesRequest<'_>) -> Result<AnthropicMessagesResponse, Error> {
     execute_messages_provider_call(prepare_messages_call(request)?).await
 }
 
-pub async fn messages_stream(request: MessagesRequest<'_>) -> CoreResult<reqwest::Response> {
+pub async fn messages_stream(request: MessagesRequest<'_>) -> Result<reqwest::Response, Error> {
     execute_messages_provider_stream(prepare_messages_call(request)?).await
 }
 
diff --git a/litellm-rust/crates/core/src/messages/prepare.rs b/litellm-rust/crates/core/src/messages/prepare.rs
index 94b5b1eaed7..3b253ac3766 100644
--- a/litellm-rust/crates/core/src/messages/prepare.rs
+++ b/litellm-rust/crates/core/src/messages/prepare.rs
@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::routing_utils::provider::{CustomLlmProvider, get_custom_llm_provider};
 
 use super::common_utils::{has_bearer_auth, has_header, messages_provider_config, string_headers};
@@ -7,7 +7,7 @@ use super::types::{MessagesRequest, ProviderMessagesRequest};
 
 pub(super) fn prepare_messages_call(
     request: MessagesRequest<'_>,
-) -> CoreResult<ProviderMessagesRequest> {
+) -> Result<ProviderMessagesRequest, Error> {
     let provider_info = get_custom_llm_provider(request.model, request.custom_llm_provider)
         .or_else(|| {
             request
@@ -18,7 +18,7 @@ pub(super) fn prepare_messages_call(
                 })
         })
         .ok_or_else(|| {
-            CoreError::InvalidProvider(
+            Error::InvalidProvider(
                 "unable to resolve custom_llm_provider for messages request".to_string(),
             )
         })?;
@@ -26,7 +26,7 @@ pub(super) fn prepare_messages_call(
     let provider = provider_info.custom_llm_provider;
 
     let config = messages_provider_config(provider)
-        .ok_or_else(|| CoreError::InvalidProvider(provider.to_string()))?;
+        .ok_or_else(|| Error::InvalidProvider(provider.to_string()))?;
     let env_lookup = |key: &str| std::env::var(key).ok();
 
     let mut headers = string_headers(request.extra_headers)?;
@@ -53,11 +53,11 @@ pub(super) fn prepare_messages_call(
 
     let url = config.complete_url(request.api_base, &model, &env_lookup)?;
     let typed_request = serde_json::from_value(request.body).map_err(|err| {
-        CoreError::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
+        Error::InvalidRequest(format!("invalid Anthropic messages request: {err}"))
     })?;
     let transformed = config.transform_request(typed_request)?;
     let body = serde_json::to_value(transformed).map_err(|err| {
-        CoreError::InvalidRequest(format!(
+        Error::InvalidRequest(format!(
             "failed to serialize Anthropic messages request: {err}"
         ))
     })?;
diff --git a/litellm-rust/crates/core/src/messages/tests.rs b/litellm-rust/crates/core/src/messages/tests.rs
index 9fc1763683b..df9f7051011 100644
--- a/litellm-rust/crates/core/src/messages/tests.rs
+++ b/litellm-rust/crates/core/src/messages/tests.rs
@@ -4,7 +4,7 @@ use serde_json::{Map, Value, json};
 use tokio::io::{AsyncReadExt, AsyncWriteExt};
 use tokio::net::{TcpListener, TcpStream};
 
-use crate::error::CoreError;
+use crate::error::Error;
 
 use super::common_utils::{
     has_bearer_auth, has_header, messages_provider_config, string_headers, truncate_error_body,
@@ -77,7 +77,7 @@ fn truncate_error_body_caps_long_payloads() {
 fn string_headers_rejects_non_string_values() {
     let headers = json!({"x-count": 3}).as_object().unwrap().clone();
     let err = string_headers(Some(headers)).expect_err("non-string header rejected");
-    assert!(matches!(err, CoreError::InvalidRequest(_)));
+    assert!(matches!(err, Error::InvalidRequest(_)));
 }
 
 #[test]
@@ -341,7 +341,7 @@ async fn messages_requires_auth_when_no_key_and_no_header() {
     .await
     .expect_err("missing auth errors");
 
-    assert!(matches!(err, CoreError::Auth(_)));
+    assert!(matches!(err, Error::Auth(_)));
 }
 
 #[tokio::test]
@@ -420,7 +420,7 @@ async fn messages_maps_provider_error_status_to_http_error() {
     .await
     .expect_err("provider error propagates");
 
-    assert!(matches!(err, CoreError::Http { status: 401, .. }));
+    assert!(matches!(err, Error::Http { status: 401, .. }));
 }
 
 #[tokio::test]
@@ -437,5 +437,5 @@ async fn messages_rejects_unsupported_provider() {
     .await
     .expect_err("unsupported provider errors");
 
-    assert!(matches!(err, CoreError::InvalidProvider(provider) if provider == "openai"));
+    assert!(matches!(err, Error::InvalidProvider(provider) if provider == "openai"));
 }
diff --git a/litellm-rust/crates/core/src/messages/transformation.rs b/litellm-rust/crates/core/src/messages/transformation.rs
index b478e20d24b..673a5728aca 100644
--- a/litellm-rust/crates/core/src/messages/transformation.rs
+++ b/litellm-rust/crates/core/src/messages/transformation.rs
@@ -1,6 +1,5 @@
-use crate::error::CoreResult;
-
 use super::types::{AnthropicMessagesRequest, AnthropicMessagesResponse};
+use crate::Error;
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
 pub enum MessagesAuthStrategy {
@@ -23,13 +22,13 @@ pub trait AnthropicMessagesProviderConfig: Sync {
         api_base: Option<&str>,
         model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn resolve_api_key(
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(&self) -> MessagesAuthStrategy {
         MessagesAuthStrategy::Header("x-api-key")
@@ -49,7 +48,7 @@ pub trait AnthropicMessagesProviderConfig: Sync {
     fn transform_request(
         &self,
         request: AnthropicMessagesRequest,
-    ) -> CoreResult<AnthropicMessagesRequest> {
+    ) -> Result<AnthropicMessagesRequest, Error> {
         Ok(request)
     }
 
@@ -57,7 +56,7 @@ pub trait AnthropicMessagesProviderConfig: Sync {
         &self,
         _model: &str,
         response: AnthropicMessagesResponse,
-    ) -> CoreResult<AnthropicMessagesResponse> {
+    ) -> Result<AnthropicMessagesResponse, Error> {
         Ok(response)
     }
 }
diff --git a/litellm-rust/crates/core/src/ocr/transformation.rs b/litellm-rust/crates/core/src/ocr/transformation.rs
index cb3e735e533..3d3c16c8cb6 100644
--- a/litellm-rust/crates/core/src/ocr/transformation.rs
+++ b/litellm-rust/crates/core/src/ocr/transformation.rs
@@ -1,7 +1,6 @@
+use crate::Error;
 use serde_json::{Map, Value};
 
-use crate::CoreResult;
-
 use super::types::{OcrRequestData, OcrResponseData};
 
 #[derive(Clone, Copy, Debug, PartialEq, Eq)]
@@ -43,13 +42,13 @@ pub trait OcrProviderConfig: Sync {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData>;
+    ) -> Result<OcrRequestData, Error>;
 
     fn transform_ocr_response(
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData>;
+    ) -> Result<OcrResponseData, Error>;
 
     fn complete_url(
         &self,
@@ -57,13 +56,13 @@ pub trait OcrProviderConfig: Sync {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn resolve_api_key(
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String>;
+    ) -> Result<String, Error>;
 
     fn auth_strategy(&self) -> OcrAuthStrategy {
         OcrAuthStrategy::Bearer
diff --git a/litellm-rust/crates/core/src/providers/anthropic/chat_completions/tests.rs b/litellm-rust/crates/core/src/providers/anthropic/chat_completions/tests.rs
index 4534ac0182c..b22de6c47de 100644
--- a/litellm-rust/crates/core/src/providers/anthropic/chat_completions/tests.rs
+++ b/litellm-rust/crates/core/src/providers/anthropic/chat_completions/tests.rs
@@ -1,4 +1,5 @@
 use super::*;
+use crate::Error;
 use serde_json::json;
 
 fn messages(value: Value) -> Vec<ChatMessage> {
@@ -19,7 +20,7 @@ fn transform(model: &str, msgs: Value, opts: Value) -> Value {
         .body
 }
 
-fn transform_response(body: Value) -> CoreResult<ChatCompletionsResponse> {
+fn transform_response(body: Value) -> Result<ChatCompletionsResponse, Error> {
     ANTHROPIC_CHAT_COMPLETIONS_CONFIG
         .transform_response("claude-sonnet-4-5", ProviderChatResponseData { body })
 }
@@ -390,29 +391,26 @@ fn declines_a_response_carrying_a_non_text_block() {
         "usage": {"input_tokens": 1, "output_tokens": 1}
     }))
     .expect_err("non-text block");
-    assert_eq!(
-        err,
-        CoreError::Unsupported("non-text response content block")
-    );
+    assert_eq!(err, Error::Unsupported("non-text response content block"));
 }
 
 #[test]
 fn errors_on_a_response_missing_required_fields() {
     assert_eq!(
         transform_response(json!("nope")).expect_err("not an object"),
-        CoreError::InvalidResponse("messages response is not an object".to_string())
+        Error::InvalidResponse("messages response is not an object".to_string())
     );
     assert_eq!(
         transform_response(json!({"model": "m", "usage": {}})).expect_err("no content"),
-        CoreError::MissingField("content")
+        Error::MissingField("content")
     );
     assert_eq!(
         transform_response(json!({"model": "m", "content": []})).expect_err("no usage"),
-        CoreError::MissingField("usage")
+        Error::MissingField("usage")
     );
     assert_eq!(
         transform_response(json!({"content": [], "usage": {}})).expect_err("no model"),
-        CoreError::MissingField("model")
+        Error::MissingField("model")
     );
 }
 
diff --git a/litellm-rust/crates/core/src/providers/anthropic/chat_completions/transformation.rs b/litellm-rust/crates/core/src/providers/anthropic/chat_completions/transformation.rs
index 3658642b539..97cc48aa6f2 100644
--- a/litellm-rust/crates/core/src/providers/anthropic/chat_completions/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/anthropic/chat_completions/transformation.rs
@@ -10,7 +10,7 @@ use crate::chat_completions::types::{
     ProviderChatRequestData, ProviderChatResponseData,
 };
 use crate::constants::ANTHROPIC_OAUTH_TOKEN_PREFIX;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::providers::anthropic::messages::transformation::{
     complete_anthropic_url, resolve_anthropic_api_key,
 };
@@ -74,7 +74,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_anthropic_url(api_base, env_lookup))
     }
 
@@ -84,7 +84,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth> {
+    ) -> Result<ChatCompletionsAuth, Error> {
         Ok(ChatCompletionsAuth::Header {
             name: "x-api-key",
             value: resolve_anthropic_api_key(api_key, env_lookup)?,
@@ -137,7 +137,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData> {
+    ) -> Result<ProviderChatRequestData, Error> {
         Ok(ProviderChatRequestData {
             body: anthropic_body(model, &build_conversation(&messages), optional_params),
         })
@@ -147,15 +147,16 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         &self,
         _model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse> {
-        let body = response.body.as_object().ok_or_else(|| {
-            CoreError::InvalidResponse("messages response is not an object".into())
-        })?;
+    ) -> Result<ChatCompletionsResponse, Error> {
+        let body = response
+            .body
+            .as_object()
+            .ok_or_else(|| Error::InvalidResponse("messages response is not an object".into()))?;
 
         let content = body
             .get("content")
             .and_then(Value::as_array)
-            .ok_or(CoreError::MissingField("content"))?;
+            .ok_or(Error::MissingField("content"))?;
         // The route declines tool and thinking requests, so a non-text block
         // means the response carries something this path never asked for.
         // Decline rather than silently dropping it; the host falls back.
@@ -163,7 +164,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
             .iter()
             .any(|block| block.get("type").and_then(Value::as_str) != Some("text"))
         {
-            return Err(CoreError::Unsupported("non-text response content block"));
+            return Err(Error::Unsupported("non-text response content block"));
         }
         let text: String = content
             .iter()
@@ -173,7 +174,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
         let usage = body
             .get("usage")
             .and_then(Value::as_object)
-            .ok_or(CoreError::MissingField("usage"))?;
+            .ok_or(Error::MissingField("usage"))?;
         let field = |name: &str| usage.get(name).and_then(Value::as_u64).unwrap_or(0);
 
         Ok(ChatCompletionsResponse {
@@ -181,7 +182,7 @@ impl ChatCompletionsProviderConfig for AnthropicChatCompletionsConfig {
             model: body
                 .get("model")
                 .and_then(Value::as_str)
-                .ok_or(CoreError::MissingField("model"))?
+                .ok_or(Error::MissingField("model"))?
                 .to_string(),
             choices: vec![ChatCompletionsChoice {
                 index: 0,
diff --git a/litellm-rust/crates/core/src/providers/anthropic/messages/transformation.rs b/litellm-rust/crates/core/src/providers/anthropic/messages/transformation.rs
index 829f2260d3c..8fcc0f36c7d 100644
--- a/litellm-rust/crates/core/src/providers/anthropic/messages/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/anthropic/messages/transformation.rs
@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::messages::transformation::{AnthropicMessagesProviderConfig, MessagesAuthStrategy};
 
 const ANTHROPIC_API_KEY_ENV: &str = "ANTHROPIC_API_KEY";
@@ -17,12 +17,12 @@ pub fn non_empty(value: Option<&str>) -> Option<&str> {
 pub fn resolve_anthropic_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(api_key)
         .map(str::to_string)
         .or_else(|| env_lookup(ANTHROPIC_API_KEY_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Anthropic API Key - Set `api_key` or the ANTHROPIC_API_KEY \
                  environment variable"
                     .to_string(),
@@ -52,7 +52,7 @@ impl AnthropicMessagesProviderConfig for AnthropicMessagesConfig {
         api_base: Option<&str>,
         _model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_anthropic_url(api_base, env_lookup))
     }
 
@@ -60,7 +60,7 @@ impl AnthropicMessagesProviderConfig for AnthropicMessagesConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_anthropic_api_key(api_key, env_lookup)
     }
 
@@ -121,7 +121,7 @@ mod tests {
         );
         assert!(matches!(
             resolve_anthropic_api_key(None, &|_| None).expect_err("missing key"),
-            CoreError::Auth(_)
+            Error::Auth(_)
         ));
     }
 
diff --git a/litellm-rust/crates/core/src/providers/azure_ai/messages/transformation.rs b/litellm-rust/crates/core/src/providers/azure_ai/messages/transformation.rs
index 7b958c77ba3..70dad0300f1 100644
--- a/litellm-rust/crates/core/src/providers/azure_ai/messages/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/azure_ai/messages/transformation.rs
@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use crate::messages::transformation::{AnthropicMessagesProviderConfig, MessagesAuthStrategy};
 use crate::messages::types::{
     AnthropicMessage, AnthropicMessagesRequest, AnthropicMessagesResponse, ContentBlock,
@@ -28,12 +28,12 @@ pub const AZURE_ANTHROPIC_MESSAGES_CONFIG: AzureAnthropicMessagesConfig =
 pub fn resolve_azure_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(api_key)
         .map(str::to_string)
         .or_else(|| env_lookup(AZURE_API_KEY_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Azure API Key - Set `api_key` or the AZURE_API_KEY environment variable"
                     .to_string(),
             )
@@ -43,12 +43,12 @@ pub fn resolve_azure_api_key(
 pub fn complete_azure_anthropic_url(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let api_base = non_empty(api_base)
         .map(str::to_string)
         .or_else(|| env_lookup(AZURE_API_BASE_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Azure API Base - Set `api_base` or the AZURE_API_BASE environment variable. \
                  Expected format: https://<resource-name>.services.ai.azure.com/anthropic"
                     .to_string(),
@@ -147,7 +147,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         api_base: Option<&str>,
         _model: &str,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_azure_anthropic_url(api_base, env_lookup)
     }
 
@@ -155,7 +155,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_azure_api_key(api_key, env_lookup)
     }
 
@@ -174,7 +174,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
     fn transform_request(
         &self,
         request: AnthropicMessagesRequest,
-    ) -> CoreResult<AnthropicMessagesRequest> {
+    ) -> Result<AnthropicMessagesRequest, Error> {
         let mut request = fold_system_role_messages(request);
         if let Some(system) = request.system.as_mut() {
             strip_scope_from_system(system);
@@ -190,7 +190,7 @@ impl AnthropicMessagesProviderConfig for AzureAnthropicMessagesConfig {
         &self,
         model: &str,
         response: AnthropicMessagesResponse,
-    ) -> CoreResult<AnthropicMessagesResponse> {
+    ) -> Result<AnthropicMessagesResponse, Error> {
         self.anthropic.transform_response(model, response)
     }
 }
@@ -268,7 +268,7 @@ mod tests {
             "https://env.services.ai.azure.com/anthropic/v1/messages"
         );
         let err = complete_azure_anthropic_url(Some("  "), &|_| None).expect_err("missing base");
-        assert!(matches!(err, CoreError::Auth(_)));
+        assert!(matches!(err, Error::Auth(_)));
     }
 
     #[test]
@@ -284,7 +284,7 @@ mod tests {
         );
         assert!(matches!(
             resolve_azure_api_key(None, &|_| None).expect_err("missing key"),
-            CoreError::Auth(_)
+            Error::Auth(_)
         ));
     }
 
diff --git a/litellm-rust/crates/core/src/providers/azure_ai/ocr/transformation.rs b/litellm-rust/crates/core/src/providers/azure_ai/ocr/transformation.rs
index eabd15677cc..b26a7925e8a 100644
--- a/litellm-rust/crates/core/src/providers/azure_ai/ocr/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/azure_ai/ocr/transformation.rs
@@ -1,6 +1,6 @@
 use std::collections::BTreeSet;
 
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::{OcrAuthStrategy, OcrProviderConfig, OcrResponseHandling};
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value, json};
@@ -32,17 +32,17 @@ fn resolve_value(
     env_name: &str,
     env_lookup: &dyn Fn(&str) -> Option<String>,
     missing_message: &str,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     non_empty(explicit)
         .map(str::to_string)
         .or_else(|| env_lookup(env_name).filter(|value| !value.trim().is_empty()))
-        .ok_or_else(|| CoreError::Auth(missing_message.to_string()))
+        .ok_or_else(|| Error::Auth(missing_message.to_string()))
 }
 
 pub fn resolve_azure_ai_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_key,
         AZURE_AI_API_KEY_ENV,
@@ -54,7 +54,7 @@ pub fn resolve_azure_ai_api_key(
 pub fn resolve_azure_ai_api_base(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_base,
         AZURE_AI_API_BASE_ENV,
@@ -66,7 +66,7 @@ pub fn resolve_azure_ai_api_base(
 pub fn complete_azure_ai_url(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let base = resolve_azure_ai_api_base(api_base, env_lookup)?;
     Ok(format!(
         "{}/providers/mistral/azure/ocr",
@@ -77,7 +77,7 @@ pub fn complete_azure_ai_url(
 pub fn resolve_document_intelligence_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_key,
         AZURE_DOCUMENT_INTELLIGENCE_API_KEY_ENV,
@@ -89,7 +89,7 @@ pub fn resolve_document_intelligence_api_key(
 pub fn resolve_document_intelligence_endpoint(
     api_base: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     resolve_value(
         api_base,
         AZURE_DOCUMENT_INTELLIGENCE_ENDPOINT_ENV,
@@ -127,7 +127,7 @@ fn pages_token_is_valid(token: &str) -> bool {
     }
 }
 
-fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
+fn normalize_pages_param(pages: &Value) -> Result<Option<String>, Error> {
     match pages {
         Value::String(value) => {
             let normalized = value
@@ -138,7 +138,7 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
             if normalized.split(',').all(pages_token_is_valid) {
                 Ok(Some(normalized))
             } else {
-                Err(CoreError::InvalidRequest(format!(
+                Err(Error::InvalidRequest(format!(
                     "Invalid `pages` string for Azure Document Intelligence: {value:?}. Expected format like '1-3,5,7-9'."
                 )))
             }
@@ -152,7 +152,7 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
                 for value in values {
                     let page = value.as_i64().expect("checked is_i64");
                     if page < 0 {
-                        return Err(CoreError::InvalidRequest(
+                        return Err(Error::InvalidRequest(
                             "`pages` integers must be >= 0 (Mistral 0-based indices)".to_string(),
                         ));
                     }
@@ -176,16 +176,16 @@ fn normalize_pages_param(pages: &Value) -> CoreResult<Option<String>> {
                 if normalized.split(',').all(pages_token_is_valid) {
                     return Ok(Some(normalized));
                 }
-                return Err(CoreError::InvalidRequest(format!(
+                return Err(Error::InvalidRequest(format!(
                     "Invalid `pages` list for Azure Document Intelligence: {values:?}. Expected tokens like '1' or '3-5'."
                 )));
             }
-            Err(CoreError::InvalidRequest(
+            Err(Error::InvalidRequest(
                 "`pages` must be a list[int] (0-based, Mistral-style) or a string like '1-3,5,7-9'."
                     .to_string(),
             ))
         }
-        _ => Err(CoreError::InvalidRequest(
+        _ => Err(Error::InvalidRequest(
             "`pages` must be a list[int] (0-based, Mistral-style) or a string like '1-3,5,7-9'."
                 .to_string(),
         )),
@@ -197,7 +197,7 @@ pub fn complete_document_intelligence_url(
     model: &str,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let endpoint = resolve_document_intelligence_endpoint(api_base, env_lookup)?;
     let mut url = format!(
         "{}/documentintelligence/documentModels/{}:analyze?api-version={}",
@@ -216,20 +216,20 @@ pub fn complete_document_intelligence_url(
     Ok(url)
 }
 
-fn document_url_from_mistral_document(document: &Value) -> CoreResult<&str> {
-    let object = document.as_object().ok_or_else(|| CoreError::InvalidType {
+fn document_url_from_mistral_document(document: &Value) -> Result<&str, Error> {
+    let object = document.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(document),
     })?;
     let doc_type = object
         .get("type")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("document.type"))?;
+        .ok_or(Error::MissingField("document.type"))?;
     let field_name = match doc_type {
         "document_url" => "document_url",
         "image_url" => "image_url",
         other => {
-            return Err(CoreError::InvalidRequest(format!(
+            return Err(Error::InvalidRequest(format!(
                 "Invalid document type: {other}. Must be 'document_url' or 'image_url'"
             )));
         }
@@ -238,7 +238,7 @@ fn document_url_from_mistral_document(document: &Value) -> CoreResult<&str> {
         .get(field_name)
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField(field_name))
+        .ok_or(Error::MissingField(field_name))
 }
 
 fn extract_base64_from_data_uri(data_uri: &str) -> &str {
@@ -290,7 +290,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
     }
 
@@ -298,7 +298,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
     }
 
@@ -308,7 +308,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_azure_ai_url(api_base, env_lookup)
     }
 
@@ -316,7 +316,7 @@ impl OcrProviderConfig for AzureAiOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_azure_ai_api_key(api_key, env_lookup)
     }
 
@@ -335,7 +335,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         _model: &str,
         document: Value,
         _optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         let document_url = document_url_from_mistral_document(&document)?;
         let mut data = Map::new();
         if document_url.starts_with("data:") {
@@ -359,19 +359,19 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
         let status = response
             .get("status")
             .and_then(Value::as_str)
-            .ok_or(CoreError::MissingField("status"))?;
+            .ok_or(Error::MissingField("status"))?;
         if status != "succeeded" {
-            return Err(CoreError::InvalidResponse(format!(
+            return Err(Error::InvalidResponse(format!(
                 "Azure Document Intelligence analysis failed with status: {status}"
             )));
         }
@@ -414,7 +414,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_document_intelligence_url(api_base, model, optional_params, env_lookup)
     }
 
@@ -422,7 +422,7 @@ impl OcrProviderConfig for AzureDocumentIntelligenceOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_document_intelligence_api_key(api_key, env_lookup)
     }
 
diff --git a/litellm-rust/crates/core/src/providers/bedrock/audio_transcription.rs b/litellm-rust/crates/core/src/providers/bedrock/audio_transcription.rs
index 5e885734182..bb4f6afe5f9 100644
--- a/litellm-rust/crates/core/src/providers/bedrock/audio_transcription.rs
+++ b/litellm-rust/crates/core/src/providers/bedrock/audio_transcription.rs
@@ -6,7 +6,7 @@ use crate::audio_transcription::transformation::{
 use crate::audio_transcription::types::{
     AudioTranscriptionRequestData, AudioTranscriptionResponseData,
 };
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 
 pub use super::aws_base::{aws_auth_config, bedrock_model_id_and_region, resolve_bedrock_region};
 use super::constants::{BEDROCK_RUNTIME_ENDPOINT_TEMPLATE, BEDROCK_SERVICE};
@@ -18,8 +18,8 @@ pub static BEDROCK_AUDIO_TRANSCRIPTION_CONFIG: BedrockAudioTranscriptionConfig =
 
 pub struct BedrockAudioTranscriptionConfig;
 
-fn audio_fields(audio: Value) -> CoreResult<(String, String)> {
-    let object = audio.as_object().ok_or_else(|| CoreError::InvalidType {
+fn audio_fields(audio: Value) -> Result<(String, String), Error> {
+    let object = audio.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(&audio),
     })?;
@@ -27,13 +27,13 @@ fn audio_fields(audio: Value) -> CoreResult<(String, String)> {
         .get("data")
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField("audio.data"))?;
+        .ok_or(Error::MissingField("audio.data"))?;
     let format = object
         .get("format")
         .and_then(Value::as_str)
         .filter(|value| matches!(*value, "wav" | "mp3" | "flac" | "ogg"))
         .ok_or_else(|| {
-            CoreError::InvalidRequest("audio.format must be wav, mp3, flac, or ogg".to_string())
+            Error::InvalidRequest("audio.format must be wav, mp3, flac, or ogg".to_string())
         })?;
     Ok((data.to_string(), format.to_string()))
 }
@@ -55,7 +55,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         _model: &str,
         audio: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<AudioTranscriptionRequestData> {
+    ) -> Result<AudioTranscriptionRequestData, Error> {
         let (data, format) = audio_fields(audio)?;
         let mut instruction = "Transcribe the audio. Respond with only the transcript.".to_string();
         if let Some(language) = optional_string(&optional_params, "language") {
@@ -87,14 +87,14 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         &self,
         _model: &str,
         response_json: Value,
-    ) -> CoreResult<AudioTranscriptionResponseData> {
+    ) -> Result<AudioTranscriptionResponseData, Error> {
         let content = response_json
             .get("output")
             .and_then(|value| value.get("message"))
             .and_then(|value| value.get("content"))
             .and_then(Value::as_array)
             .ok_or_else(|| {
-                CoreError::InvalidResponse("Bedrock response has no output content".to_string())
+                Error::InvalidResponse("Bedrock response has no output content".to_string())
             })?;
         let mut text = String::new();
         for block in content {
@@ -111,7 +111,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         let (model_id, model_region) = bedrock_model_id_and_region(model);
         let region = resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup);
         let endpoint = optional_params
@@ -133,7 +133,7 @@ impl AudioTranscriptionProviderConfig for BedrockAudioTranscriptionConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<AudioTranscriptionAuth> {
+    ) -> Result<AudioTranscriptionAuth, Error> {
         let (_, model_region) = bedrock_model_id_and_region(model);
         Ok(AudioTranscriptionAuth::AwsSigV4 {
             region: resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup),
diff --git a/litellm-rust/crates/core/src/providers/bedrock/aws_base.rs b/litellm-rust/crates/core/src/providers/bedrock/aws_base.rs
index b11639aa09b..e5e52bfce95 100644
--- a/litellm-rust/crates/core/src/providers/bedrock/aws_base.rs
+++ b/litellm-rust/crates/core/src/providers/bedrock/aws_base.rs
@@ -4,7 +4,7 @@ use std::time::Duration;
 use std::time::{SystemTime, UNIX_EPOCH};
 
 use crate::caching::in_memory_cache::InMemoryCache;
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 use aws_credential_types::Credentials;
 use aws_credential_types::provider::ProvideCredentials;
 use aws_sigv4::http_request::{
@@ -197,7 +197,7 @@ pub fn classify_auth(
 pub async fn resolve_credentials(
     config: AwsAuthConfig,
     env_lookup: &(dyn Fn(&str) -> Option<String> + Sync),
-) -> CoreResult<Credentials> {
+) -> Result<Credentials, Error> {
     let resolved = config.clone().with_environment(env_lookup);
     let flow = classify_auth(config, env_lookup);
     match flow {
@@ -244,9 +244,10 @@ pub async fn resolve_credentials(
             let provider = aws_config::profile::ProfileFileCredentialsProvider::builder()
                 .profile_name(name)
                 .build();
-            provider.provide_credentials().await.map_err(|error| {
-                CoreError::Auth(format!("AWS profile credentials failed: {error}"))
-            })
+            provider
+                .provide_credentials()
+                .await
+                .map_err(|error| Error::Auth(format!("AWS profile credentials failed: {error}")))
         }
         AwsAuthFlow::AssumeRole { role, session_name } => {
             if is_already_running_as_role(&role, &resolved).await? {
@@ -260,7 +261,7 @@ pub async fn resolve_credentials(
                         .build()
                         .await;
                 let credentials = provider.provide_credentials().await.map_err(|error| {
-                    CoreError::Auth(format!("AWS default credentials failed: {error}"))
+                    Error::Auth(format!("AWS default credentials failed: {error}"))
                 })?;
                 set_cached_credentials(
                     key,
@@ -301,7 +302,7 @@ pub async fn resolve_credentials(
             provider
                 .provide_credentials()
                 .await
-                .map_err(|error| CoreError::Auth(format!("AWS role credentials failed: {error}")))
+                .map_err(|error| Error::Auth(format!("AWS role credentials failed: {error}")))
         }
         AwsAuthFlow::WebIdentity {
             token,
@@ -325,13 +326,13 @@ pub async fn resolve_credentials(
                 .send()
                 .await
                 .map_err(|error| {
-                    CoreError::Auth(format!("AWS web identity credentials failed: {error}"))
+                    Error::Auth(format!("AWS web identity credentials failed: {error}"))
                 })?;
             let credentials = response.credentials().ok_or_else(|| {
-                CoreError::Auth("AWS web identity response had no credentials".to_string())
+                Error::Auth("AWS web identity response had no credentials".to_string())
             })?;
             let expiration = SystemTime::try_from(*credentials.expiration()).map_err(|error| {
-                CoreError::Auth(format!("AWS web identity expiration was invalid: {error}"))
+                Error::Auth(format!("AWS web identity expiration was invalid: {error}"))
             })?;
             Ok(Credentials::new(
                 credentials.access_key_id(),
@@ -350,9 +351,10 @@ pub async fn resolve_credentials(
                 aws_config::default_provider::credentials::DefaultCredentialsChain::builder()
                     .build()
                     .await;
-            let credentials = provider.provide_credentials().await.map_err(|error| {
-                CoreError::Auth(format!("AWS default credentials failed: {error}"))
-            })?;
+            let credentials = provider
+                .provide_credentials()
+                .await
+                .map_err(|error| Error::Auth(format!("AWS default credentials failed: {error}")))?;
             set_cached_credentials(
                 key,
                 credentials.clone(),
@@ -363,7 +365,7 @@ pub async fn resolve_credentials(
     }
 }
 
-async fn is_already_running_as_role(role: &str, config: &AwsAuthConfig) -> CoreResult<bool> {
+async fn is_already_running_as_role(role: &str, config: &AwsAuthConfig) -> Result<bool, Error> {
     if role_identity(role).is_none() {
         return Ok(false);
     }
@@ -437,7 +439,7 @@ pub fn sign_bedrock_post(
     region: &str,
     credentials: &Credentials,
     signing_time: SystemTime,
-) -> CoreResult<BTreeMap<String, String>> {
+) -> Result<BTreeMap<String, String>, Error> {
     let identity: Identity = credentials.clone().into();
     let params = v4::SigningParams::builder()
         .identity(&identity)
@@ -447,14 +449,14 @@ pub fn sign_bedrock_post(
         .settings(SigningSettings::default())
         .build()
         .map(SigningParams::from)
-        .map_err(|error| CoreError::Auth(format!("AWS signing parameters failed: {error}")))?;
+        .map_err(|error| Error::Auth(format!("AWS signing parameters failed: {error}")))?;
     let header_refs = headers
         .iter()
         .map(|(name, value)| (name.as_str(), value.as_str()));
     let request = SignableRequest::new("POST", url, header_refs, SignableBody::Bytes(body))
-        .map_err(|error| CoreError::Auth(format!("AWS signable request failed: {error}")))?;
+        .map_err(|error| Error::Auth(format!("AWS signable request failed: {error}")))?;
     let (instructions, _) = sign(request, &params)
-        .map_err(|error| CoreError::Auth(format!("AWS request signing failed: {error}")))?
+        .map_err(|error| Error::Auth(format!("AWS request signing failed: {error}")))?
         .into_parts();
     Ok(instructions
         .headers()
diff --git a/litellm-rust/crates/core/src/providers/bedrock/chat_completions/tests.rs b/litellm-rust/crates/core/src/providers/bedrock/chat_completions/tests.rs
index 4b75dcb8e9d..c86f061b9ca 100644
--- a/litellm-rust/crates/core/src/providers/bedrock/chat_completions/tests.rs
+++ b/litellm-rust/crates/core/src/providers/bedrock/chat_completions/tests.rs
@@ -1,4 +1,5 @@
 use super::*;
+use crate::Error;
 use serde_json::json;
 
 fn messages(value: Value) -> Vec<ChatMessage> {
@@ -23,7 +24,7 @@ fn transform(msgs: Value, opts: Value) -> Value {
         .body
 }
 
-fn transform_response(body: Value) -> CoreResult<ChatCompletionsResponse> {
+fn transform_response(body: Value) -> Result<ChatCompletionsResponse, Error> {
     BEDROCK_CHAT_COMPLETIONS_CONFIG.transform_response(
         "anthropic.claude-sonnet-4-5-v1:0",
         ProviderChatResponseData { body },
@@ -478,25 +479,22 @@ fn declines_a_response_carrying_a_tool_use_block() {
         "usage": {"inputTokens": 1, "outputTokens": 1}
     }))
     .expect_err("tool use block");
-    assert_eq!(
-        err,
-        CoreError::Unsupported("non-text response content block")
-    );
+    assert_eq!(err, Error::Unsupported("non-text response content block"));
 }
 
 #[test]
 fn errors_on_a_response_missing_required_fields() {
     assert_eq!(
         transform_response(json!("nope")).expect_err("not an object"),
-        CoreError::InvalidResponse("converse response is not an object".to_string())
+        Error::InvalidResponse("converse response is not an object".to_string())
     );
     assert_eq!(
         transform_response(json!({"usage": {}})).expect_err("no output"),
-        CoreError::MissingField("output.message.content")
+        Error::MissingField("output.message.content")
     );
     assert_eq!(
         transform_response(json!({"output": {"message": {"content": []}}})).expect_err("no usage"),
-        CoreError::MissingField("usage")
+        Error::MissingField("usage")
     );
 }
 
diff --git a/litellm-rust/crates/core/src/providers/bedrock/chat_completions/transformation.rs b/litellm-rust/crates/core/src/providers/bedrock/chat_completions/transformation.rs
index b107950748e..ef5f44b4a14 100644
--- a/litellm-rust/crates/core/src/providers/bedrock/chat_completions/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/bedrock/chat_completions/transformation.rs
@@ -11,7 +11,7 @@ use crate::chat_completions::types::{
     ChatCompletionsUsage, ChatMessage, ChatMessageContent, ProviderChatRequestData,
     ProviderChatResponseData,
 };
-use crate::error::{CoreError, CoreResult};
+use crate::error::Error;
 
 use super::super::aws_base::{bedrock_model_id_and_region, resolve_bedrock_region};
 use super::super::constants::{AWS_BEARER_TOKEN_BEDROCK, BEDROCK_RUNTIME_ENDPOINT_TEMPLATE};
@@ -110,7 +110,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         let (model_id, model_region) = bedrock_model_id_and_region(model);
         let region = resolve_bedrock_region(model_region.as_deref(), optional_params, env_lookup);
         let endpoint = optional_params
@@ -137,7 +137,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<ChatCompletionsAuth> {
+    ) -> Result<ChatCompletionsAuth, Error> {
         // Python reads `api_key` as the Bedrock bearer token and consults the
         // env only when the caller passed none, so a caller-supplied empty key
         // falls through to SigV4 without reaching for the environment. An
@@ -208,7 +208,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         _model: &str,
         messages: Vec<ChatMessage>,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<ProviderChatRequestData> {
+    ) -> Result<ProviderChatRequestData, Error> {
         Ok(ProviderChatRequestData {
             body: converse_body(&build_conversation(&messages), &optional_params),
         })
@@ -218,17 +218,18 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         &self,
         model: &str,
         response: ProviderChatResponseData,
-    ) -> CoreResult<ChatCompletionsResponse> {
-        let body = response.body.as_object().ok_or_else(|| {
-            CoreError::InvalidResponse("converse response is not an object".into())
-        })?;
+    ) -> Result<ChatCompletionsResponse, Error> {
+        let body = response
+            .body
+            .as_object()
+            .ok_or_else(|| Error::InvalidResponse("converse response is not an object".into()))?;
 
         let content = body
             .get("output")
             .and_then(|output| output.get("message"))
             .and_then(|message| message.get("content"))
             .and_then(Value::as_array)
-            .ok_or(CoreError::MissingField("output.message.content"))?;
+            .ok_or(Error::MissingField("output.message.content"))?;
         // The route declines tool requests, so anything other than a text block
         // is something this path never asked for. Decline; the host falls back.
         if content.iter().any(|block| {
@@ -236,7 +237,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
                 .as_object()
                 .is_none_or(|block| block.len() != 1 || !block.contains_key("text"))
         }) {
-            return Err(CoreError::Unsupported("non-text response content block"));
+            return Err(Error::Unsupported("non-text response content block"));
         }
         let text: String = content
             .iter()
@@ -246,7 +247,7 @@ impl ChatCompletionsProviderConfig for BedrockChatCompletionsConfig {
         let usage = body
             .get("usage")
             .and_then(Value::as_object)
-            .ok_or(CoreError::MissingField("usage"))?;
+            .ok_or(Error::MissingField("usage"))?;
         let field = |name: &str| usage.get(name).and_then(Value::as_u64).unwrap_or(0);
         let computed = usage_from_parts(
             field("inputTokens"),
diff --git a/litellm-rust/crates/core/src/providers/mistral/ocr/transformation.rs b/litellm-rust/crates/core/src/providers/mistral/ocr/transformation.rs
index dc720cc4244..6a8a38204a9 100644
--- a/litellm-rust/crates/core/src/providers/mistral/ocr/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/mistral/ocr/transformation.rs
@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::OcrProviderConfig;
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value};
@@ -47,7 +47,7 @@ pub fn complete_url(api_base: Option<&str>) -> String {
 
 /// Resolve the Mistral API key from the explicit param or the environment.
 ///
-/// Blank/whitespace values are treated as absent. Returns `CoreError::Auth`
+/// Blank/whitespace values are treated as absent. Returns `Error::Auth`
 /// when no usable key is available.
 ///
 /// Note: the env fallback only reads the process environment. Secret-manager
@@ -56,13 +56,13 @@ pub fn complete_url(api_base: Option<&str>) -> String {
 pub fn resolve_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
         .map(str::to_string)
         .or_else(|| env_lookup(MISTRAL_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
-        .ok_or_else(|| CoreError::Auth(MISSING_KEY_MESSAGE.to_string()))
+        .ok_or_else(|| Error::Auth(MISSING_KEY_MESSAGE.to_string()))
 }
 
 pub struct MistralOcrConfig;
@@ -79,9 +79,9 @@ impl OcrProviderConfig for MistralOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         if !document.is_object() {
-            return Err(CoreError::InvalidType {
+            return Err(Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&document),
             });
@@ -104,10 +104,10 @@ impl OcrProviderConfig for MistralOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response_object = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
@@ -140,7 +140,7 @@ impl OcrProviderConfig for MistralOcrConfig {
         _model: &str,
         _optional_params: &Map<String, Value>,
         _env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         Ok(complete_url(api_base))
     }
 
@@ -148,7 +148,7 @@ impl OcrProviderConfig for MistralOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_api_key(api_key, env_lookup)
     }
 }
@@ -165,11 +165,11 @@ pub fn transform_ocr_request(
     model: &str,
     document: Value,
     optional_params: Map<String, Value>,
-) -> CoreResult<OcrRequestData> {
+) -> Result<OcrRequestData, Error> {
     MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
 }
 
-pub fn transform_ocr_response(model: &str, response_json: Value) -> CoreResult<OcrResponseData> {
+pub fn transform_ocr_response(model: &str, response_json: Value) -> Result<OcrResponseData, Error> {
     MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
 }
 
@@ -250,7 +250,7 @@ mod tests {
 
         assert_eq!(
             err,
-            CoreError::InvalidType {
+            Error::InvalidType {
                 expected: "object",
                 actual: "string",
             }
@@ -307,6 +307,6 @@ mod tests {
     #[test]
     fn resolve_api_key_errors_when_absent() {
         let err = resolve_api_key(None, &|_| None).expect_err("missing key should error");
-        assert_eq!(err, CoreError::Auth(MISSING_KEY_MESSAGE.to_string()));
+        assert_eq!(err, Error::Auth(MISSING_KEY_MESSAGE.to_string()));
     }
 }
diff --git a/litellm-rust/crates/core/src/providers/openai/realtime/transformation.rs b/litellm-rust/crates/core/src/providers/openai/realtime/transformation.rs
index b3f6b03b28a..f1985f81b7d 100644
--- a/litellm-rust/crates/core/src/providers/openai/realtime/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/openai/realtime/transformation.rs
@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::realtime::transformation::RealtimeProviderConfig;
 use crate::realtime::types::{RealtimeEvent, RealtimeTransformResult};
 
@@ -72,7 +72,7 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
         &self,
         event: &RealtimeEvent,
         _model: &str,
-    ) -> CoreResult<RealtimeTransformResult> {
+    ) -> Result<RealtimeTransformResult, Error> {
         Ok(RealtimeTransformResult::passthrough(event.clone()))
     }
 
@@ -80,7 +80,7 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
         &self,
         event: &RealtimeEvent,
         _model: &str,
-    ) -> CoreResult<RealtimeTransformResult> {
+    ) -> Result<RealtimeTransformResult, Error> {
         Ok(RealtimeTransformResult::passthrough(event.clone()))
     }
 }
@@ -88,14 +88,14 @@ impl RealtimeProviderConfig for OpenAiRealtimeConfig {
 pub fn transform_realtime_request(
     event: &RealtimeEvent,
     model: &str,
-) -> CoreResult<RealtimeTransformResult> {
+) -> Result<RealtimeTransformResult, Error> {
     OPENAI_REALTIME_CONFIG.transform_realtime_request(event, model)
 }
 
 pub fn transform_realtime_response(
     event: &RealtimeEvent,
     model: &str,
-) -> CoreResult<RealtimeTransformResult> {
+) -> Result<RealtimeTransformResult, Error> {
     OPENAI_REALTIME_CONFIG.transform_realtime_response(event, model)
 }
 
diff --git a/litellm-rust/crates/core/src/providers/openai/responses/transformation.rs b/litellm-rust/crates/core/src/providers/openai/responses/transformation.rs
index e15197c468c..be86bb90311 100644
--- a/litellm-rust/crates/core/src/providers/openai/responses/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/openai/responses/transformation.rs
@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsTransformResult};
 use crate::responses::websocket::{ResponsesWebSocketProviderConfig, enforce_model};
 
@@ -15,7 +15,7 @@ impl ResponsesWebSocketProviderConfig for OpenAIResponsesWsConfig {
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult> {
+    ) -> Result<ResponsesWsTransformResult, Error> {
         Ok(ResponsesWsTransformResult::passthrough(enforce_model(
             event, model,
         )))
@@ -25,7 +25,7 @@ impl ResponsesWebSocketProviderConfig for OpenAIResponsesWsConfig {
         &self,
         event: &ResponsesWsEvent,
         _model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult> {
+    ) -> Result<ResponsesWsTransformResult, Error> {
         Ok(ResponsesWsTransformResult::passthrough(event.clone()))
     }
 }
diff --git a/litellm-rust/crates/core/src/providers/vertex_ai/ocr/transformation.rs b/litellm-rust/crates/core/src/providers/vertex_ai/ocr/transformation.rs
index 6300149c237..ee095447028 100644
--- a/litellm-rust/crates/core/src/providers/vertex_ai/ocr/transformation.rs
+++ b/litellm-rust/crates/core/src/providers/vertex_ai/ocr/transformation.rs
@@ -1,4 +1,4 @@
-use crate::error::{CoreError, CoreResult, json_type_name};
+use crate::error::{Error, json_type_name};
 use crate::ocr::transformation::OcrProviderConfig;
 use crate::ocr::types::{OcrRequestData, OcrResponseData};
 use serde_json::{Map, Value, json};
@@ -43,7 +43,7 @@ pub fn is_deepseek_model(model: &str) -> bool {
 pub fn resolve_vertex_api_key(
     api_key: Option<&str>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     api_key
         .map(str::trim)
         .filter(|key| !key.is_empty())
@@ -51,7 +51,7 @@ pub fn resolve_vertex_api_key(
         .or_else(|| env_lookup(VERTEX_AI_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
         .or_else(|| env_lookup(VERTEXAI_API_KEY_ENV).filter(|key| !key.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::Auth(
+            Error::Auth(
                 "Missing Vertex AI access token - pass api_key or provide Authorization via extra_headers"
                     .to_string(),
             )
@@ -61,12 +61,12 @@ pub fn resolve_vertex_api_key(
 fn vertex_project(
     params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     string_param(params, &["vertex_project", "vertex_ai_project"])
         .map(str::to_string)
         .or_else(|| env_lookup(VERTEXAI_PROJECT_ENV).filter(|value| !value.trim().is_empty()))
         .ok_or_else(|| {
-            CoreError::InvalidRequest(
+            Error::InvalidRequest(
                 "Missing vertex_project - Set VERTEXAI_PROJECT environment variable or pass vertex_project parameter"
                     .to_string(),
             )
@@ -99,7 +99,7 @@ pub fn complete_vertex_mistral_url(
     model: &str,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let project = vertex_project(optional_params, env_lookup)?;
     let location = vertex_location(optional_params, env_lookup);
     let base = vertex_mistral_api_base(api_base, &location);
@@ -112,7 +112,7 @@ pub fn complete_vertex_deepseek_url(
     api_base: Option<&str>,
     optional_params: &Map<String, Value>,
     env_lookup: &dyn Fn(&str) -> Option<String>,
-) -> CoreResult<String> {
+) -> Result<String, Error> {
     let project = vertex_project(optional_params, env_lookup)?;
     let location = vertex_location(optional_params, env_lookup);
     let base = api_base
@@ -125,20 +125,20 @@ pub fn complete_vertex_deepseek_url(
     ))
 }
 
-fn document_content_item(document: &Value) -> CoreResult<Value> {
-    let object = document.as_object().ok_or_else(|| CoreError::InvalidType {
+fn document_content_item(document: &Value) -> Result<Value, Error> {
+    let object = document.as_object().ok_or_else(|| Error::InvalidType {
         expected: "object",
         actual: json_type_name(document),
     })?;
     let doc_type = object
         .get("type")
         .and_then(Value::as_str)
-        .ok_or(CoreError::MissingField("document.type"))?;
+        .ok_or(Error::MissingField("document.type"))?;
     let url_field = match doc_type {
         "image_url" => "image_url",
         "document_url" => "document_url",
         other => {
-            return Err(CoreError::InvalidRequest(format!(
+            return Err(Error::InvalidRequest(format!(
                 "Unsupported document type: {other}. Expected 'image_url' or 'document_url'"
             )));
         }
@@ -147,7 +147,7 @@ fn document_content_item(document: &Value) -> CoreResult<Value> {
         .get(url_field)
         .and_then(Value::as_str)
         .filter(|value| !value.is_empty())
-        .ok_or(CoreError::MissingField(url_field))?;
+        .ok_or(Error::MissingField(url_field))?;
 
     Ok(json!({
         "type": "image_url",
@@ -163,7 +163,7 @@ fn deepseek_model_name(model: &str) -> String {
     }
 }
 
-fn first_choice_content(response: &Value) -> CoreResult<Value> {
+fn first_choice_content(response: &Value) -> Result<Value, Error> {
     response
         .get("choices")
         .and_then(Value::as_array)
@@ -176,9 +176,7 @@ fn first_choice_content(response: &Value) -> CoreResult<Value> {
             Value::Object(_) => true,
             _ => false,
         })
-        .ok_or_else(|| {
-            CoreError::InvalidResponse("No content in DeepSeek OCR response".to_string())
-        })
+        .ok_or_else(|| Error::InvalidResponse("No content in DeepSeek OCR response".to_string()))
 }
 
 fn ocr_data_from_content(content: Value, usage: Option<Value>, model: &str) -> Value {
@@ -219,7 +217,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_request(model, document, optional_params)
     }
 
@@ -227,7 +225,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         MISTRAL_OCR_CONFIG.transform_ocr_response(model, response_json)
     }
 
@@ -237,7 +235,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_vertex_mistral_url(api_base, model, optional_params, env_lookup)
     }
 
@@ -245,7 +243,7 @@ impl OcrProviderConfig for VertexAiOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_vertex_api_key(api_key, env_lookup)
     }
 
@@ -264,7 +262,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         model: &str,
         document: Value,
         optional_params: Map<String, Value>,
-    ) -> CoreResult<OcrRequestData> {
+    ) -> Result<OcrRequestData, Error> {
         let mut data = Map::new();
         data.insert(
             "model".to_string(),
@@ -289,10 +287,10 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         &self,
         model: &str,
         response_json: Value,
-    ) -> CoreResult<OcrResponseData> {
+    ) -> Result<OcrResponseData, Error> {
         let response = response_json
             .as_object()
-            .ok_or_else(|| CoreError::InvalidType {
+            .ok_or_else(|| Error::InvalidType {
                 expected: "object",
                 actual: json_type_name(&response_json),
             })?;
@@ -314,7 +312,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
             });
         }
 
-        let object = ocr_data.as_object().ok_or_else(|| CoreError::InvalidType {
+        let object = ocr_data.as_object().ok_or_else(|| Error::InvalidType {
             expected: "object",
             actual: json_type_name(&ocr_data),
         })?;
@@ -346,7 +344,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         _model: &str,
         optional_params: &Map<String, Value>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         complete_vertex_deepseek_url(api_base, optional_params, env_lookup)
     }
 
@@ -354,7 +352,7 @@ impl OcrProviderConfig for VertexAiDeepSeekOcrConfig {
         &self,
         api_key: Option<&str>,
         env_lookup: &dyn Fn(&str) -> Option<String>,
-    ) -> CoreResult<String> {
+    ) -> Result<String, Error> {
         resolve_vertex_api_key(api_key, env_lookup)
     }
 }
diff --git a/litellm-rust/crates/core/src/realtime/transformation.rs b/litellm-rust/crates/core/src/realtime/transformation.rs
index 69b88687000..b08084514ef 100644
--- a/litellm-rust/crates/core/src/realtime/transformation.rs
+++ b/litellm-rust/crates/core/src/realtime/transformation.rs
@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::realtime::types::{RealtimeEvent, RealtimeTransformResult};
 
 pub trait RealtimeProviderConfig {
@@ -11,12 +11,12 @@ pub trait RealtimeProviderConfig {
         &self,
         event: &RealtimeEvent,
         model: &str,
-    ) -> CoreResult<RealtimeTransformResult>;
+    ) -> Result<RealtimeTransformResult, Error>;
 
     /// Transform a backend → client event before it is forwarded downstream.
     fn transform_realtime_response(
         &self,
         event: &RealtimeEvent,
         model: &str,
-    ) -> CoreResult<RealtimeTransformResult>;
+    ) -> Result<RealtimeTransformResult, Error>;
 }
diff --git a/litellm-rust/crates/core/src/responses/instrumentation.rs b/litellm-rust/crates/core/src/responses/instrumentation.rs
index ec04571da14..b1098f4d386 100644
--- a/litellm-rust/crates/core/src/responses/instrumentation.rs
+++ b/litellm-rust/crates/core/src/responses/instrumentation.rs
@@ -5,9 +5,9 @@ use std::time::{SystemTime, UNIX_EPOCH};
 
 use serde_json::Value;
 
+use crate::Error;
 use crate::call_lifecycle::{CallLifecycleContext, CallLifecycleHooks, CallLifecycleTiming};
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsEventType};
-use crate::{CoreError, CoreResult};
 
 #[derive(Clone, Debug, Default, PartialEq, Eq)]
 pub struct ResponsesWsUsage {
@@ -205,7 +205,7 @@ impl ResponsesWsInstrumentation {
     }
 }
 
-type LifecycleFuture<'a, T> = Pin<Box<dyn Future<Output = CoreResult<T>> + Send + 'a>>;
+type LifecycleFuture<'a, T> = Pin<Box<dyn Future<Output = Result<T, Error>> + Send + 'a>>;
 
 impl CallLifecycleHooks<(), (), ()> for ResponsesWsInstrumentation {
     type PreCallFuture<'a> = LifecycleFuture<'a, ()>;
@@ -246,7 +246,7 @@ impl CallLifecycleHooks<(), (), ()> for ResponsesWsInstrumentation {
     fn async_log_failure_event<'a>(
         &'a self,
         _context: &'a CallLifecycleContext,
-        _error: &'a CoreError,
+        _error: &'a Error,
         _timing: &'a CallLifecycleTiming,
     ) -> Self::FailureFuture<'a> {
         Box::pin(async move {
@@ -342,7 +342,7 @@ mod tests {
                 ),
                 (),
                 &instrumentation,
-                |_| async { Ok::<(), CoreError>(()) },
+                |_| async { Ok::<(), Error>(()) },
             )
             .await;
 
diff --git a/litellm-rust/crates/core/src/responses/websocket.rs b/litellm-rust/crates/core/src/responses/websocket.rs
index 92dc19627a0..5d037e9cf1b 100644
--- a/litellm-rust/crates/core/src/responses/websocket.rs
+++ b/litellm-rust/crates/core/src/responses/websocket.rs
@@ -1,4 +1,4 @@
-use crate::CoreResult;
+use crate::Error;
 use crate::constants::{OPENAI_RESPONSES_DEFAULT_API_BASE, OPENAI_RESPONSES_PATH};
 use crate::responses::types::{ResponsesWsEvent, ResponsesWsEventType, ResponsesWsTransformResult};
 
@@ -19,13 +19,13 @@ pub trait ResponsesWebSocketProviderConfig: Sync {
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult>;
+    ) -> Result<ResponsesWsTransformResult, Error>;
 
     fn transform_ws_response(
         &self,
         event: &ResponsesWsEvent,
         model: &str,
-    ) -> CoreResult<ResponsesWsTransformResult>;
+    ) -> Result<ResponsesWsTransformResult, Error>;
 }
 
 pub fn complete_websocket_url(
diff --git a/litellm-rust/crates/core/tests/workspace_crate_allowlist.rs b/litellm-rust/crates/core/tests/workspace_crate_allowlist.rs
index 656ba033b62..8a8a5ea263a 100644
--- a/litellm-rust/crates/core/tests/workspace_crate_allowlist.rs
+++ b/litellm-rust/crates/core/tests/workspace_crate_allowlist.rs
@@ -1,7 +1,8 @@
-//! Enforcement: the litellm-rust workspace has exactly three crates.
+//! Enforcement: the litellm-rust workspace has exactly four crates.
 //!
-//! `core` (pure translation), `ai-gateway` (routes + all network I/O), and
-//! `python-bridge` (the PyO3 cdylib). Adding or removing a crate must be a
+//! `core` (the Rust SDK), `ai-gateway` (the HTTP/WebSocket host),
+//! `python-interop` (domain-neutral PyO3 primitives), and `python-bridge` (the
+//! PyO3 cdylib). Adding or removing a crate must be a
 //! deliberate act: this test fails until the allowlist here is updated, forcing
 //! whoever changes the crate set to justify the new crate per the rule that a
 //! crate is a layer needing independent compilation / its own deps / a separate
@@ -16,10 +17,15 @@ use std::path::{Path, PathBuf};
 
 /// The one true crate set. Update BOTH this and `litellm-rust/AGENTS.md` when the
 /// workspace legitimately gains or loses a crate.
-const EXPECTED_MEMBERS: &[&str] = &["crates/core", "crates/ai-gateway", "crates/python-bridge"];
+const EXPECTED_MEMBERS: &[&str] = &[
+    "crates/core",
+    "crates/ai-gateway",
+    "crates/python-interop",
+    "crates/python-bridge",
+];
 
 /// The crate subdirectory names that must exist under `crates/`.
-const EXPECTED_CRATE_DIRS: &[&str] = &["core", "ai-gateway", "python-bridge"];
+const EXPECTED_CRATE_DIRS: &[&str] = &["core", "ai-gateway", "python-interop", "python-bridge"];
 
 const MISMATCH: &str = "litellm-rust crate set changed — update this allowlist AND litellm-rust/AGENTS.md, and justify the crate per the rule (crate = layer needing independent compilation / its own deps / a separate artifact).";
 
diff --git a/litellm-rust/crates/python-bridge/AGENTS.md b/litellm-rust/crates/python-bridge/AGENTS.md
index ad3cddfa5fd..42282ca4da4 100644
--- a/litellm-rust/crates/python-bridge/AGENTS.md
+++ b/litellm-rust/crates/python-bridge/AGENTS.md
@@ -1,3 +1,3 @@
-litellm-python-bridge is the PyO3 cdylib that exposes Rust to the litellm Python SDK — a thin adapter (Python objects → Rust calls → Python results) over the litellm-core route entrypoints (e.g. `litellm_core::messages::messages`).
+litellm-python-bridge is the PyO3 cdylib that exposes LiteLLM Rust APIs to the Python SDK. Keep API registration, domain dependency wiring, request assembly, and Python exception mapping here. Put domain-neutral Python/Serde conversion and GIL primitives in litellm-python-interop.
 
 Keep it thin: no business logic, no transforms, no I/O orchestration — just marshal in/out and call the core entrypoint.
diff --git a/litellm-rust/crates/python-bridge/CLAUDE.md b/litellm-rust/crates/python-bridge/CLAUDE.md
index 3ce8b8c639a..d25ae5a8130 100644
--- a/litellm-rust/crates/python-bridge/CLAUDE.md
+++ b/litellm-rust/crates/python-bridge/CLAUDE.md
@@ -5,8 +5,9 @@ Rules for `litellm-rust/crates/python-bridge`.
 ## Responsibility
 
 `python-bridge` is the PyO3 boundary between Python LiteLLM and Rust transforms.
-Keep this crate thin. It adapts Python objects to Rust payloads and returns
-Python-compatible dictionaries.
+Keep this crate thin. It exposes LiteLLM Rust APIs, assembles domain requests,
+maps domain errors to Python exceptions, and delegates generic conversion and
+GIL handling to `litellm-python-interop`.
 
 ## Bridge Shape
 
diff --git a/litellm-rust/crates/python-bridge/Cargo.toml b/litellm-rust/crates/python-bridge/Cargo.toml
index d461a483ae0..498003de149 100644
--- a/litellm-rust/crates/python-bridge/Cargo.toml
+++ b/litellm-rust/crates/python-bridge/Cargo.toml
@@ -13,14 +13,14 @@ crate-type = ["cdylib"]
 default = ["abi3"]
 abi3 = ["pyo3/abi3-py310"]
 extension-module = ["pyo3/extension-module"]
+panic-test = []
 
 [dependencies]
 litellm-core = { workspace = true, features = ["bedrock-auth"] }
 litellm-ai-gateway = { workspace = true, default-features = false }
+litellm-python-interop.workspace = true
 pyo3.workspace = true
 pyo3-async-runtimes.workspace = true
-pythonize.workspace = true
-serde.workspace = true
 serde_json.workspace = true
 tokio.workspace = true
 
diff --git a/litellm-rust/crates/python-bridge/benches/serialization.rs b/litellm-rust/crates/python-bridge/benches/serialization.rs
index 8a90cf667d0..0b9436d0cb7 100644
--- a/litellm-rust/crates/python-bridge/benches/serialization.rs
+++ b/litellm-rust/crates/python-bridge/benches/serialization.rs
@@ -2,6 +2,7 @@ use std::hint::black_box;
 use std::time::Duration;
 
 use criterion::{BenchmarkId, Criterion, criterion_group, criterion_main};
+use litellm_python_interop::{from_py, to_py};
 use pyo3::prelude::*;
 use pyo3::types::PyDict;
 use serde_json::{Value, json};
@@ -25,7 +26,7 @@ fn former_json_roundtrip_from_py(py: Python<'_>, value: &Bound<'_, PyAny>) -> Va
 }
 
 fn pythonize_from_py(value: &Bound<'_, PyAny>) -> Value {
-    pythonize::depythonize(value).expect("payload should depythonize")
+    from_py(value).expect("payload should depythonize")
 }
 
 fn former_json_roundtrip_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
@@ -37,12 +38,10 @@ fn former_json_roundtrip_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
 }
 
 fn pythonize_to_py(py: Python<'_>, value: &Value) -> Py<PyAny> {
-    pythonize::pythonize(py, value)
-        .expect("response should pythonize")
-        .unbind()
+    to_py(py, value).expect("response should pythonize")
 }
 
-fn serialization(c: &mut Criterion) {
+fn bridge_serialization(c: &mut Criterion) {
     Python::initialize();
     Python::attach(|py| {
         for &(label, payload_bytes) in PAYLOAD_SIZES {
@@ -98,6 +97,6 @@ criterion_group! {
         .sample_size(20)
         .warm_up_time(Duration::from_secs(1))
         .measurement_time(Duration::from_secs(4));
-    targets = serialization
+    targets = bridge_serialization
 }
 criterion_main!(benches);
diff --git a/litellm-rust/crates/python-bridge/src/gil.rs b/litellm-rust/crates/python-bridge/src/gil.rs
deleted file mode 100644
index e887c8ec1e3..00000000000
--- a/litellm-rust/crates/python-bridge/src/gil.rs
+++ /dev/null
@@ -1,32 +0,0 @@
-//! GIL accounting.
-//!
-//! A single chokepoint for releasing the GIL around blocking work. Every
-//! blocking call in the bridge goes through [`release_gil`] instead of calling
-//! `Python::detach` directly, so the release count stays accurate and we
-//! have one place to extend later (timing histograms, per-call labels, etc.).
-
-use std::sync::atomic::{AtomicU64, Ordering};
-
-use pyo3::prelude::*;
-
-/// Number of times the bridge has released the GIL since process start.
-static GIL_RELEASES: AtomicU64 = AtomicU64::new(0);
-
-/// Release the GIL around `f`, recording the release.
-///
-/// `f` must not touch any Python state — that is what makes releasing the GIL
-/// safe. Returning the value back to Python re-acquires the GIL at the call
-/// site, after `f` has finished.
-pub fn release_gil<T, F>(py: Python<'_>, f: F) -> T
-where
-    F: FnOnce() -> T + Send,
-    T: Send,
-{
-    GIL_RELEASES.fetch_add(1, Ordering::Relaxed);
-    py.detach(f)
-}
-
-/// Total GIL releases performed by the bridge so far.
-pub fn release_count() -> u64 {
-    GIL_RELEASES.load(Ordering::Relaxed)
-}
diff --git a/litellm-rust/crates/python-bridge/src/lib.rs b/litellm-rust/crates/python-bridge/src/lib.rs
index f9e75f45f75..68aa9436b15 100644
--- a/litellm-rust/crates/python-bridge/src/lib.rs
+++ b/litellm-rust/crates/python-bridge/src/lib.rs
@@ -10,19 +10,15 @@ use litellm_core::chat_completions::types::{ChatCompletionsRequest, ChatCompleti
 use litellm_core::chat_completions::{
     chat_completions as run_chat_completions, chat_completions_decline_reason,
 };
-use litellm_core::error::CoreError;
+use litellm_core::error::Error;
 use litellm_core::messages::messages as run_messages;
 use litellm_core::messages::types::{AnthropicMessagesResponse, MessagesRequest};
+use litellm_python_interop::{from_py, release_count, release_gil, to_py};
 use pyo3::exceptions::{PyRuntimeError, PyValueError};
 use pyo3::prelude::*;
 use pyo3::types::{PyAny, PyDict};
 use serde_json::{Map, Value};
 
-mod gil;
-mod marshal;
-
-use marshal::{from_py, to_py};
-
 pyo3::create_exception!(
     _native,
     RustBridgeDeclined,
@@ -58,13 +54,13 @@ fn chat_completions_response_to_py(
     to_py(py, &response)
 }
 
-fn core_error_to_pyerr(err: CoreError) -> PyErr {
+fn core_error_to_pyerr(err: Error) -> PyErr {
     match err {
-        CoreError::Auth(message) => PyValueError::new_err(message),
-        CoreError::InvalidProvider(_)
-        | CoreError::InvalidRequest(_)
-        | CoreError::InvalidType { .. }
-        | CoreError::MissingField(_) => PyValueError::new_err(err.to_string()),
+        Error::Auth(message) => PyValueError::new_err(message),
+        Error::InvalidProvider(_)
+        | Error::InvalidRequest(_)
+        | Error::InvalidType { .. }
+        | Error::MissingField(_) => PyValueError::new_err(err.to_string()),
         other => PyRuntimeError::new_err(other.to_string()),
     }
 }
@@ -75,22 +71,22 @@ fn core_error_to_pyerr(err: CoreError) -> PyErr {
 /// Everything raised before the request goes out is safe for the host to retry
 /// on its own path; anything after it is not, because the provider has already
 /// done the work and billed for it.
-fn chat_completions_error_to_pyerr(err: CoreError) -> PyErr {
+fn chat_completions_error_to_pyerr(err: Error) -> PyErr {
     match err {
-        CoreError::Unsupported(_)
-        | CoreError::Auth(_)
-        | CoreError::InvalidProvider(_)
-        | CoreError::InvalidRequest(_)
-        | CoreError::InvalidType { .. }
-        | CoreError::MissingField(_)
-        | CoreError::Routing(_)
+        Error::Unsupported(_)
+        | Error::Auth(_)
+        | Error::InvalidProvider(_)
+        | Error::InvalidRequest(_)
+        | Error::InvalidType { .. }
+        | Error::MissingField(_)
+        | Error::Routing(_)
         // Nothing reached the provider, so serving it on Python cannot double
         // bill and is the only way the caller gets an answer at all.
-        | CoreError::Connect(_) => RustBridgeDeclined::new_err(err.to_string()),
-        CoreError::Http { status, body } => {
+        | Error::Connect(_) => RustBridgeDeclined::new_err(err.to_string()),
+        Error::Http { status, body } => {
             RustUpstreamError::new_err((status, format!("{status}: {body}")))
         }
-        CoreError::Network(message) | CoreError::InvalidResponse(message) => {
+        Error::Network(message) | Error::InvalidResponse(message) => {
             RustUpstreamError::new_err((0u16, message))
         }
     }
@@ -230,7 +226,7 @@ fn ocr(
         timeout_seconds,
     )?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_ocr(OcrRequest {
             model: &model,
             document,
@@ -318,7 +314,7 @@ fn transcription(
     };
     let optional_params = optional_object_to_map(py, "optional_params", optional_params)?;
     let timeout = optional_timeout(timeout_seconds);
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_audio_transcription(
             AudioTranscriptionRequest {
                 model: &model,
@@ -419,7 +415,7 @@ fn messages(
     let (body, extra_headers, timeout) =
         marshal_messages_inputs(py, body, extra_headers, timeout_seconds)?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_messages(MessagesRequest {
             model: &model,
             body,
@@ -546,7 +542,7 @@ fn chat_completions(
         timeout_seconds,
     )?;
 
-    let result = gil::release_gil(py, || {
+    let result = release_gil(py, || {
         pyo3_async_runtimes::tokio::get_runtime().block_on(run_chat_completions(
             ChatCompletionsRequest {
                 model: &model,
@@ -610,10 +606,16 @@ fn achat_completions(
 #[pyfunction]
 fn gil_stats(py: Python<'_>) -> PyResult<Py<PyAny>> {
     let stats = PyDict::new(py);
-    stats.set_item("releases", gil::release_count())?;
+    stats.set_item("releases", release_count())?;
     Ok(stats.into_any().unbind())
 }
 
+#[cfg(feature = "panic-test")]
+#[pyfunction]
+fn _panic_for_test() {
+    panic!("intentional PyO3 panic smoke test");
+}
+
 #[pymodule]
 fn _native(module: &Bound<'_, PyModule>) -> PyResult<()> {
     let py = module.py();
@@ -630,5 +632,7 @@ fn _native(module: &Bound<'_, PyModule>) -> PyResult<()> {
     module.add_function(wrap_pyfunction!(achat_completions, module)?)?;
     module.add_class::<ResponsesWebSocketConnection>()?;
     module.add_function(wrap_pyfunction!(gil_stats, module)?)?;
+    #[cfg(feature = "panic-test")]
+    module.add_function(wrap_pyfunction!(_panic_for_test, module)?)?;
     Ok(())
 }
diff --git a/litellm-rust/crates/python-bridge/tests/marshal_boundary.rs b/litellm-rust/crates/python-bridge/tests/marshal_boundary.rs
index 6a6ede22e85..d397d20b9fd 100644
--- a/litellm-rust/crates/python-bridge/tests/marshal_boundary.rs
+++ b/litellm-rust/crates/python-bridge/tests/marshal_boundary.rs
@@ -1,7 +1,7 @@
 use std::fs;
 use std::path::{Path, PathBuf};
 
-const DISALLOWED_OUTSIDE_MARSHAL: &[&str] = &[
+const DISALLOWED_OUTSIDE_INTEROP: &[&str] = &[
     "py.import(\"json\")",
     "pythonize::",
     "serde_json::to_string",
@@ -33,18 +33,15 @@ fn rust_sources(directory: &Path) -> Vec<PathBuf> {
 }
 
 #[test]
-fn serialization_is_centralized_in_marshal_module() {
+fn serialization_uses_the_interop_boundary() {
     let root = source_root();
 
     for path in rust_sources(&root) {
-        if path == root.join("marshal.rs") {
-            continue;
-        }
         let source = fs::read_to_string(&path).expect("bridge source should be readable");
-        for disallowed in DISALLOWED_OUTSIDE_MARSHAL {
+        for disallowed in DISALLOWED_OUTSIDE_INTEROP {
             assert!(
                 !source.contains(disallowed),
-                "{} bypasses the typed marshal module with `{disallowed}`",
+                "{} bypasses litellm-python-interop with `{disallowed}`",
                 path.display()
             );
         }
diff --git a/litellm-rust/crates/python-interop/AGENTS.md b/litellm-rust/crates/python-interop/AGENTS.md
new file mode 100644
index 00000000000..d1d61e5dfa0
--- /dev/null
+++ b/litellm-rust/crates/python-interop/AGENTS.md
@@ -0,0 +1 @@
+litellm-python-interop is the domain-neutral PyO3 foundation. Keep generic Python/Serde conversion and interpreter primitives here. Do not add LiteLLM domain crates, route types, API registration, or cdylib build features.
diff --git a/litellm-rust/crates/python-interop/Cargo.toml b/litellm-rust/crates/python-interop/Cargo.toml
new file mode 100644
index 00000000000..9da6af6e2e2
--- /dev/null
+++ b/litellm-rust/crates/python-interop/Cargo.toml
@@ -0,0 +1,15 @@
+[package]
+name = "litellm-python-interop"
+version = "0.1.0"
+edition.workspace = true
+license.workspace = true
+repository.workspace = true
+
+[dependencies]
+pyo3.workspace = true
+pythonize.workspace = true
+serde.workspace = true
+
+[dev-dependencies]
+rstest.workspace = true
+serde_json.workspace = true
diff --git a/litellm-rust/crates/python-interop/src/gil.rs b/litellm-rust/crates/python-interop/src/gil.rs
new file mode 100644
index 00000000000..04b966a6002
--- /dev/null
+++ b/litellm-rust/crates/python-interop/src/gil.rs
@@ -0,0 +1,21 @@
+use std::sync::atomic::{AtomicU64, Ordering};
+
+use pyo3::prelude::*;
+
+static GIL_RELEASES: AtomicU64 = AtomicU64::new(0);
+
+/// Runs work detached from the interpreter and records the release.
+///
+/// `f` must not access Python state while the interpreter is detached.
+pub fn release_gil<T, F>(py: Python<'_>, f: F) -> T
+where
+    F: FnOnce() -> T + Send,
+    T: Send,
+{
+    GIL_RELEASES.fetch_add(1, Ordering::Relaxed);
+    py.detach(f)
+}
+
+pub fn release_count() -> u64 {
+    GIL_RELEASES.load(Ordering::Relaxed)
+}
diff --git a/litellm-rust/crates/python-interop/src/lib.rs b/litellm-rust/crates/python-interop/src/lib.rs
new file mode 100644
index 00000000000..df2bd260fdb
--- /dev/null
+++ b/litellm-rust/crates/python-interop/src/lib.rs
@@ -0,0 +1,5 @@
+mod gil;
+mod marshal;
+
+pub use gil::{release_count, release_gil};
+pub use marshal::{from_py, to_py};
diff --git a/litellm-rust/crates/python-bridge/src/marshal.rs b/litellm-rust/crates/python-interop/src/marshal.rs
similarity index 100%
rename from litellm-rust/crates/python-bridge/src/marshal.rs
rename to litellm-rust/crates/python-interop/src/marshal.rs
diff --git a/litellm-rust/crates/python-interop/tests/interop.rs b/litellm-rust/crates/python-interop/tests/interop.rs
new file mode 100644
index 00000000000..9c456dcb938
--- /dev/null
+++ b/litellm-rust/crates/python-interop/tests/interop.rs
@@ -0,0 +1,44 @@
+use pyo3::Python;
+use rstest::{fixture, rstest};
+use serde_json::{Value, json};
+
+use litellm_python_interop::{from_py, release_count, release_gil, to_py};
+
+struct InitializedPython;
+
+impl InitializedPython {
+    fn attach<F, R>(&self, f: F) -> R
+    where
+        F: for<'py> FnOnce(Python<'py>) -> R,
+    {
+        Python::attach(f)
+    }
+}
+
+#[fixture]
+#[once]
+fn initialized_python() -> InitializedPython {
+    Python::initialize();
+    InitializedPython
+}
+
+#[rstest]
+fn serde_values_round_trip_through_python(#[from(initialized_python)] python: &InitializedPython) {
+    python.attach(|py| {
+        let expected = json!({"model": "test", "items": [1, true, null]});
+        let python_value = to_py(py, &expected).expect("value should convert to Python");
+        let actual: Value =
+            from_py(python_value.bind(py)).expect("Python value should convert to serde");
+
+        assert_eq!(actual, expected);
+    });
+}
+
+#[rstest]
+fn release_gil_runs_work_and_records_it(#[from(initialized_python)] python: &InitializedPython) {
+    let before = release_count();
+    let result = python.attach(|py| release_gil(py, || 42));
+
+    assert_eq!(result, 42);
+    assert_eq!(release_count(), before + 1);
+}
diff --git a/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py b/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py
index 32252711997..1e8cc4ff90e 100644
--- a/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py
+++ b/litellm/a2a_protocol/providers/bedrock_agentcore/transformation.py
@@ -10,8 +10,19 @@ from collections.abc import AsyncIterator, Mapping
 from typing import Any, Final
 
 from litellm._logging import verbose_logger
+from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+    A2A_USER_API_KEY_HASH_PARAM,
+)
+from litellm.a2a_protocol.utils import (
+    get_session_id_from_a2a_params,
+    scope_session_to_principal,
+)
+from litellm.exceptions import BadRequestError
 from litellm.llms.bedrock.chat.agentcore.transformation import AmazonAgentCoreConfig
 
+RUNTIME_SESSION_ID_MIN_LENGTH: Final = 33
+RUNTIME_SESSION_ID_MAX_LENGTH: Final = 256
+
 # Reserved outbound header names that must never be sourced from per-request
 # ``agent_extra_headers`` for AgentCore requests. ``agent_extra_headers`` carries
 # values rewritten from the client-controlled ``x-a2a-{agent}-*`` convention, so
@@ -19,8 +30,9 @@ from litellm.llms.bedrock.chat.agentcore.transformation import AmazonAgentCoreCo
 # request identity / SigV4 metadata by overwriting headers the proxy sets from
 # trusted server-side config.
 #
-# The runtime headers (session / user id) are derived server-side from
-# ``runtimeSessionId`` / ``runtimeUserId`` in the agent's ``litellm_params``;
+# The runtime headers (session / user id) are derived server-side from the A2A
+# ``message.contextId`` and ``runtimeSessionId`` / ``runtimeUserId`` in the
+# agent's ``litellm_params``;
 # ``authorization`` is set by the AgentCore signer (JWT or SigV4); ``host`` and
 # the ``x-amz-*`` family are owned by SigV4 itself.
 _RESERVED_EXACT_HEADERS: Final = frozenset(
@@ -66,6 +78,31 @@ def _filter_reserved_headers(
     return filtered or None
 
 
+def _request_scoped_runtime_session_id(
+    params: Mapping[str, Any],
+    litellm_params: Mapping[str, Any],
+) -> str | None:
+    context_id: Final = get_session_id_from_a2a_params(params)
+    if not isinstance(context_id, str) or not context_id:
+        return None
+    return scope_session_to_principal(context_id, litellm_params.get(A2A_USER_API_KEY_HASH_PARAM))
+
+
+def _validate_runtime_session_id(session_id: str, model: str) -> str:
+    if RUNTIME_SESSION_ID_MIN_LENGTH <= len(session_id) <= RUNTIME_SESSION_ID_MAX_LENGTH:
+        return session_id
+    raise BadRequestError(
+        message=(
+            f"Invalid AgentCore runtime session id {session_id!r}: AWS requires "
+            f"{RUNTIME_SESSION_ID_MIN_LENGTH}-{RUNTIME_SESSION_ID_MAX_LENGTH} characters. It is built from the A2A "
+            "message.contextId (prefixed with a 16-hex-char hash of the calling key and '-') when set, "
+            "otherwise from the agent's configured runtimeSessionId."
+        ),
+        model=model,
+        llm_provider="bedrock",
+    )
+
+
 class BedrockAgentCoreA2ATransformation:
     """
     Request/response transformation for Bedrock AgentCore A2A agents.
@@ -100,7 +137,9 @@ class BedrockAgentCoreA2ATransformation:
                 here to prevent a caller-controlled ``x-a2a-{agent}-*`` header from
                 spoofing the AgentCore runtime user id or other SigV4 metadata. Use
                 ``api_key`` / ``runtimeUserId`` / ``runtimeSessionId`` in litellm_params
-                (not ``agent_extra_headers``) to override those values.
+                (not ``agent_extra_headers``) to override those values. The runtime
+                session id is taken from ``params["message"]["contextId"]`` (scoped to
+                the calling key) when present, then ``runtimeSessionId``, else generated.
 
         Returns:
             Tuple of (url, signed_headers, signed_body_bytes)
@@ -139,7 +178,11 @@ class BedrockAgentCoreA2ATransformation:
         # Set required AgentCore session headers (normally set by transform_request,
         # which we skip because it also builds {"prompt": "..."})
         headers: Final[dict] = {}
-        session_id: Final = agentcore_config._get_runtime_session_id(optional_params)
+        session_id: Final = _validate_runtime_session_id(
+            _request_scoped_runtime_session_id(params, litellm_params)
+            or agentcore_config._get_runtime_session_id(optional_params),
+            model=model,
+        )
         headers["X-Amzn-Bedrock-AgentCore-Runtime-Session-Id"] = session_id
         runtime_user_id: Final = agentcore_config._get_runtime_user_id(optional_params)
         if runtime_user_id:
diff --git a/litellm/a2a_protocol/utils.py b/litellm/a2a_protocol/utils.py
index f2e61f66105..7c459daf720 100644
--- a/litellm/a2a_protocol/utils.py
+++ b/litellm/a2a_protocol/utils.py
@@ -2,6 +2,8 @@
 Utility functions for A2A protocol.
 """
 
+import hashlib
+from collections.abc import Mapping
 from typing import TYPE_CHECKING, Any, Final
 
 import litellm
@@ -140,6 +142,29 @@ class A2ARequestUtils:
         return prompt_tokens, completion_tokens, total_tokens
 
 
+def get_session_id_from_a2a_params(params: Mapping[str, Any]) -> str | None:
+    message: Final = params.get("message", {})
+    if isinstance(message, dict):
+        return message.get("contextId")
+    return getattr(message, "contextId", None)
+
+
+def scope_session_to_principal(session_id: str, principal: str | None) -> str:
+    """
+    Bind a client-supplied A2A contextId to the authenticated principal.
+
+    Without this, two distinct keys authorized for the same agent could set the
+    same contextId and read/append to each other's backend memory. The
+    principal is hashed (it is already a hashed token) so the raw value is never
+    sent to the agent backend, while the original contextId is kept as a suffix
+    for operator-side correlation.
+    """
+    if not principal:
+        return session_id
+    principal_prefix: Final = hashlib.sha256(principal.encode("utf-8")).hexdigest()[:16]
+    return f"{principal_prefix}-{session_id}"
+
+
 # Backwards compatibility aliases
 def extract_text_from_a2a_message(message: Any) -> str:
     return A2ARequestUtils.extract_text_from_message(message)
diff --git a/litellm/constants.py b/litellm/constants.py
index 1bd977dd9a9..1c1939bd350 100644
--- a/litellm/constants.py
+++ b/litellm/constants.py
@@ -9,6 +9,38 @@ DEFAULT_HEALTH_CHECK_PROMPT: Final = str(os.getenv("DEFAULT_HEALTH_CHECK_PROMPT"
 AZURE_DEFAULT_RESPONSES_API_VERSION: Final = str(os.getenv("AZURE_DEFAULT_RESPONSES_API_VERSION", "preview"))
 ROUTER_MAX_FALLBACKS: Final = int(os.getenv("ROUTER_MAX_FALLBACKS", 5))
 ROUTER_FALLBACK_ERROR_DETAIL_MAX_CHARS: Final = 2000
+RUNTIME_UPDATABLE_ROUTER_SETTINGS: Final[frozenset[str]] = frozenset(
+    {
+        "routing_strategy_args",
+        "routing_strategy",
+        "routing_groups",
+        "allowed_fails",
+        "cooldown_time",
+        "num_retries",
+        "timeout",
+        "max_retries",
+        "retry_after",
+        "fallbacks",
+        "context_window_fallbacks",
+        "retry_policy",
+        "model_group_retry_policy",
+        "model_group_alias",
+        "enable_weighted_failover",
+        "enable_tag_filtering",
+        "tag_routing_prefix",
+        "optional_pre_call_checks",
+    }
+)
+ROUTER_SETTINGS_MANAGED_OUTSIDE_CONFIG: Final[frozenset[str]] = frozenset(
+    {
+        "model_list",
+        "search_tools",
+        "assistants_config",
+        "router_general_settings",
+        "ignore_invalid_deployments",
+        "fallback_access_check",
+    }
+)
 DEFAULT_BATCH_SIZE: Final = int(os.getenv("DEFAULT_BATCH_SIZE", 512))
 DEFAULT_FLUSH_INTERVAL_SECONDS: Final = int(os.getenv("DEFAULT_FLUSH_INTERVAL_SECONDS", 5))
 DEFAULT_S3_FLUSH_INTERVAL_SECONDS: Final = int(os.getenv("DEFAULT_S3_FLUSH_INTERVAL_SECONDS", 10))
diff --git a/litellm/containers/main.py b/litellm/containers/main.py
index 97ca11872c1..90d59af009f 100644
--- a/litellm/containers/main.py
+++ b/litellm/containers/main.py
@@ -1,7 +1,7 @@
 import asyncio
 import contextvars
 import json
-from collections.abc import Coroutine, Mapping
+from collections.abc import Callable, Coroutine, Mapping
 from functools import partial
 from typing import Final, Literal, overload
 
@@ -47,6 +47,13 @@ __all__ = [
 
 
 ##### Container Create #######################
+async def _encode_created_container_id(
+    pending: Coroutine[object, object, ContainerObject],
+    encode: Callable[[ContainerObject], ContainerObject],
+) -> ContainerObject:
+    return encode(await pending)
+
+
 @client
 async def acreate_container(
     name: str,
@@ -256,16 +263,16 @@ def create_container(
             _is_async=_is_async,
         )
 
-        # Encode container_id with provider/model metadata for routing
+        encode: Final = partial(
+            ContainerRequestUtils.encode_container_id_in_response,
+            custom_llm_provider=custom_llm_provider,
+            litellm_metadata=kwargs.get("litellm_metadata"),
+            extra_body=extra_body,
+        )
         if isinstance(container_obj, ContainerObject):
-            container_obj = ContainerRequestUtils.encode_container_id_in_response(
-                response_obj=container_obj,
-                custom_llm_provider=custom_llm_provider,
-                litellm_metadata=kwargs.get("litellm_metadata"),
-                extra_body=extra_body,
-            )
+            return encode(container_obj)
 
-        return container_obj
+        return _encode_created_container_id(pending=container_obj, encode=encode)
 
     except Exception as e:
         raise litellm.exception_type(
diff --git a/litellm/integrations/custom_guardrail.py b/litellm/integrations/custom_guardrail.py
index e87ac9521ae..372c9bf6b91 100644
--- a/litellm/integrations/custom_guardrail.py
+++ b/litellm/integrations/custom_guardrail.py
@@ -1,4 +1,5 @@
 import contextvars
+import copy
 import hashlib
 import os
 import secrets
@@ -39,6 +40,7 @@ except ImportError:
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.llms.base_llm.guardrail_translation.base_translation import BaseTranslation
 dc: Final = DualCache()
 
 
@@ -852,6 +854,69 @@ class CustomGuardrail(CustomLogger):
 
         return result
 
+    async def async_logging_hook(
+        self,
+        kwargs: dict,  # mutable-ok: CustomLogger.async_logging_hook contract
+        result: object,
+        call_type: str,
+    ) -> tuple[dict, object]:  # mutable-ok: CustomLogger.async_logging_hook contract
+        """logging_only: run apply_guardrail on copies of the logged request/response and record the verdict."""
+        from litellm.llms import get_guardrail_translation_mapping
+
+        if not self.uses_apply_guardrail_interface() or self.use_native_lifecycle_hooks:
+            return kwargs, result
+        try:
+            translation: Final = get_guardrail_translation_mapping(CallTypes(call_type))()
+        except ValueError:
+            verbose_logger.debug(
+                "Guardrail %s: no guardrail translation for call_type=%s, skipping logging_only scan",
+                self.guardrail_name,
+                call_type,
+            )
+            return kwargs, result
+        litellm_params: Final = kwargs.get("litellm_params") or {}
+        scratch_metadata: Final = {
+            key: value
+            for key, value in (litellm_params.get("metadata") or {}).items()
+            if key != "standard_logging_guardrail_information"
+        }
+        try:
+            await self._scan_logged_call(kwargs, result, translation, scratch_metadata)
+        except Exception as e:
+            verbose_logger.warning("Guardrail %s: logging_only scan raised: %s", self.guardrail_name, e)
+        recorded: Final = scratch_metadata.get("standard_logging_guardrail_information")
+        standard_logging_object: Final = kwargs.get("standard_logging_object")
+        if not recorded or not isinstance(standard_logging_object, dict):
+            return kwargs, result
+        entries: Final = recorded if isinstance(recorded, list) else [recorded]
+        existing: Final = standard_logging_object.get("guardrail_information") or []
+        return {
+            **kwargs,
+            "standard_logging_object": {**standard_logging_object, "guardrail_information": [*existing, *entries]},
+        }, result
+
+    async def _scan_logged_call(
+        self,
+        kwargs: dict,  # mutable-ok: CustomLogger.async_logging_hook contract
+        result: object,
+        translation: "BaseTranslation",
+        scratch_metadata: dict,  # mutable-ok: apply_guardrail records its verdict into request metadata
+    ) -> None:
+        optional_params: Final = kwargs.get("optional_params") or {}
+        scratch_input: Final = copy.deepcopy(kwargs.get("messages") or kwargs.get("input"))
+        scratch_request: Final = {
+            "model": kwargs.get("model"),
+            "messages": scratch_input,
+            "input": scratch_input,
+            "tools": copy.deepcopy(optional_params.get("tools")),
+            "litellm_call_id": kwargs.get("litellm_call_id"),
+            "metadata": scratch_metadata,
+        }
+        await translation.process_input_messages(data=scratch_request, guardrail_to_apply=self)
+        await translation.process_output_response(
+            response=copy.deepcopy(result), guardrail_to_apply=self, request_data=scratch_request
+        )
+
     def supports_scan_only_tool_results(self) -> bool:
         """Whether this guardrail can scan tool-result content.
 
diff --git a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
index 3978a01a5db..0e01577b20e 100644
--- a/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
+++ b/litellm/litellm_core_utils/streaming_chunk_builder_utils.py
@@ -36,6 +36,8 @@ from litellm.types.utils import (
 from litellm.utils import print_verbose, token_counter
 
 if TYPE_CHECKING:
+    from openai.types.completion_usage import CompletionUsage
+
     from litellm.litellm_core_utils.litellm_logging import Logging
     from litellm.types.litellm_core_utils.streaming_chunk_builder_utils import (
         UsagePerChunk,
@@ -794,7 +796,7 @@ class ChunkProcessor:
 
     @staticmethod
     def _extract_usage_chunk(chunk: "_UsageBearingChunk | ModelResponse | ModelResponseStream") -> Usage | None:
-        usage_chunk: Usage | None = None
+        usage_chunk: Usage | CompletionUsage | None = None
         if hasattr(chunk, "usage") and chunk.usage is not None:
             usage_chunk = chunk.usage
         elif "usage" in chunk:
@@ -806,7 +808,9 @@ class ChunkProcessor:
 
         if isinstance(usage_chunk, dict):
             return Usage(**usage_chunk)
-        return usage_chunk
+        if usage_chunk is None or isinstance(usage_chunk, Usage):
+            return usage_chunk
+        return Usage(**usage_chunk.model_dump())
 
     def _calculate_usage_per_chunk(
         self,
diff --git a/litellm/llms/anthropic/common_utils.py b/litellm/llms/anthropic/common_utils.py
index c60ebd844ba..d23690976ad 100644
--- a/litellm/llms/anthropic/common_utils.py
+++ b/litellm/llms/anthropic/common_utils.py
@@ -1378,31 +1378,38 @@ def process_anthropic_headers(headers: httpx.Headers | dict) -> dict:
     return additional_headers
 
 
-def _anthropic_model_entry(model: ModelInfoResponse, created_at: str) -> Mapping[str, object]:
+def _anthropic_model_entry(
+    model: ModelInfoResponse, created_at: str, display_names: Mapping[str, str]
+) -> Mapping[str, object]:
     return {  # mutable-ok: JSON response body, serialized by the route and never mutated
         "type": "model",
         "id": model["id"],
-        "display_name": model["id"],
+        "display_name": display_names.get(model["id"], model["id"]),
         "created_at": created_at,
         "max_input_tokens": model.get("max_input_tokens"),
         "max_tokens": model.get("max_output_tokens"),
     }
 
 
-def create_anthropic_model_list_response(models: Sequence[ModelInfoResponse]) -> Mapping[str, object]:
+def create_anthropic_model_list_response(
+    models: Sequence[ModelInfoResponse],
+    display_names: Mapping[str, str] = MappingProxyType({}),
+) -> Mapping[str, object]:
     """Build the Anthropic-native /v1/models envelope.
 
     Clients that send an anthropic-version header parse the Anthropic Models API
     shape (type/display_name/created_at plus has_more/first_id/last_id) and filter
     the list themselves, so every model is returned here. The token limits carry
     over from the OpenAI-shaped listing, named as the Messages API names them, and
-    are always present because the vendor shape declares them nullable, not optional
+    are always present because the vendor shape declares them nullable, not optional.
+    display_names maps a listed model id to a configured human-readable name; ids
+    without an entry fall back to the id itself, matching the vendor behavior
     """
     created_at: Final = (
         datetime.fromtimestamp(DEFAULT_MODEL_CREATED_AT_TIME, tz=timezone.utc).isoformat().replace("+00:00", "Z")
     )
     data: Final = [  # mutable-ok: JSON response body, serialized by the route and never mutated
-        _anthropic_model_entry(model, created_at) for model in models
+        _anthropic_model_entry(model, created_at, display_names) for model in models
     ]
     return {  # mutable-ok: JSON response body, serialized by the route and never mutated
         "data": data,
diff --git a/litellm/llms/azure_ai/vector_stores/transformation.py b/litellm/llms/azure_ai/vector_stores/transformation.py
index 5e61d0a1dd9..64e72b819b1 100644
--- a/litellm/llms/azure_ai/vector_stores/transformation.py
+++ b/litellm/llms/azure_ai/vector_stores/transformation.py
@@ -19,6 +19,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     LiteLLMLoggingObj = _LiteLLMLoggingObj
 else:
@@ -115,6 +116,7 @@ class AzureAIVectorStoreConfig(BaseVectorStoreConfig, BaseAzureLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict[str, Any]]:
         """
         Transform search request for Azure AI Search API
diff --git a/litellm/llms/base_llm/vector_store/transformation.py b/litellm/llms/base_llm/vector_store/transformation.py
index 02a51a8bace..63e99c0915a 100644
--- a/litellm/llms/base_llm/vector_store/transformation.py
+++ b/litellm/llms/base_llm/vector_store/transformation.py
@@ -17,6 +17,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     from ..chat.transformation import BaseLLMException as _BaseLLMException
 
@@ -57,6 +58,7 @@ class BaseVectorStoreConfig:
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         pass
 
@@ -69,6 +71,7 @@ class BaseVectorStoreConfig:
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         """
         Optional async version of transform_search_vector_store_request.
@@ -84,6 +87,7 @@ class BaseVectorStoreConfig:
             litellm_logging_obj=litellm_logging_obj,
             litellm_params=litellm_params,
             extra_body=extra_body,
+            router=router,
         )
 
     @abstractmethod
@@ -197,6 +201,7 @@ class BaseDirectVectorStoreConfig(BaseVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: Mapping[str, object],
         extra_body: Mapping[str, object] | None = None,
+        router: "Router | None" = None,
     ) -> NoReturn:
         raise NotImplementedError("Direct vector store providers execute the search themselves; no HTTP request shape")
 
diff --git a/litellm/llms/bedrock/base_aws_llm.py b/litellm/llms/bedrock/base_aws_llm.py
index 852cfaa24f2..1e634ced29b 100644
--- a/litellm/llms/bedrock/base_aws_llm.py
+++ b/litellm/llms/bedrock/base_aws_llm.py
@@ -1442,7 +1442,7 @@ class BaseAWSLLM:
     @tracer.wrap()
     def get_request_headers(
         self,
-        credentials: Credentials,
+        credentials: Credentials | None,
         aws_region_name: str,
         extra_headers: dict | None,
         endpoint_url: str,
@@ -1469,9 +1469,13 @@ class BaseAWSLLM:
             try:
                 from botocore.auth import SigV4Auth
                 from botocore.awsrequest import AWSRequest
+                from botocore.exceptions import NoCredentialsError
             except ImportError:
                 raise ImportError("Missing boto3 to call bedrock. Run 'pip install boto3'.")
 
+            if credentials is None:
+                raise NoCredentialsError()
+
             # Filter headers for AWS signature calculation
             # AWS SigV4 only includes specific headers in signature calculation
             aws_signature_headers: Final = self._filter_headers_for_aws_signature(headers)
diff --git a/litellm/llms/bedrock/chat/converse_handler.py b/litellm/llms/bedrock/chat/converse_handler.py
index ca5f1298360..7d5f99ca893 100644
--- a/litellm/llms/bedrock/chat/converse_handler.py
+++ b/litellm/llms/bedrock/chat/converse_handler.py
@@ -1,4 +1,6 @@
 import json
+from collections.abc import Mapping
+from types import MappingProxyType
 from typing import Any, Final
 
 import httpx
@@ -24,6 +26,22 @@ from ..common_utils import BedrockError, _get_all_bedrock_regions
 from .invoke_handler import AWSEventStreamDecoder, MockResponseIterator, make_call
 
 
+def _sigv4_principal(credentials: Credentials | None) -> Mapping[str, str]:
+    if credentials is None:
+        return MappingProxyType({})
+    return MappingProxyType(
+        {
+            key: value
+            for key, value in (
+                ("aws_access_key_id", credentials.access_key),
+                ("aws_secret_access_key", credentials.secret_key),
+                ("aws_session_token", credentials.token),
+            )
+            if value is not None
+        }
+    )
+
+
 def make_sync_call(
     client: HTTPHandler | None,
     api_base: str,
@@ -95,7 +113,7 @@ class BedrockConverseLLM(BaseAWSLLM):
         stream,
         optional_params: dict,
         litellm_params: dict,
-        credentials: Credentials,
+        credentials: Credentials | None,
         logger_fn=None,
         headers={},
         client: AsyncHTTPHandler | None = None,
@@ -167,7 +185,7 @@ class BedrockConverseLLM(BaseAWSLLM):
         stream,
         optional_params: dict,
         litellm_params: dict,
-        credentials: Credentials,
+        credentials: Credentials | None,
         logger_fn=None,
         headers: dict = {},
         client: AsyncHTTPHandler | None = None,
@@ -331,7 +349,7 @@ class BedrockConverseLLM(BaseAWSLLM):
 
         litellm_params["aws_region_name"] = aws_region_name  # [DO NOT DELETE] important for async calls
 
-        credentials: Final[Credentials] = self.get_credentials(
+        credentials: Final[Credentials | None] = self.get_credentials(
             aws_access_key_id=aws_access_key_id,
             aws_secret_access_key=aws_secret_access_key,
             aws_session_token=aws_session_token,
@@ -368,19 +386,13 @@ class BedrockConverseLLM(BaseAWSLLM):
         # The Rust core owns the whole call for the subset it accepts. Ask
         # before transforming so whichever path runs emits pre_call once, and
         # hand down the credentials, region and endpoint this handler already
-        # resolved so both paths sign as the same principal.
+        # resolved so both paths sign as the same principal. Bearer-token auth
+        # resolves no SigV4 principal at all, and each path reads that token
+        # itself.
         rust_optional_params: Final = {  # mutable-ok: json.dumps in the bridge rejects a mappingproxy
             **optional_params,
-            **{  # mutable-ok: merged into its mutable parent above
-                key: value
-                for key, value in (
-                    ("aws_access_key_id", credentials.access_key),
-                    ("aws_secret_access_key", credentials.secret_key),
-                    ("aws_session_token", credentials.token),
-                    ("aws_region_name", aws_region_name),
-                )
-                if value is not None
-            },
+            **_sigv4_principal(credentials),
+            "aws_region_name": aws_region_name,
         }
         serves_via_rust: Final = rust_chat_completions_accepts(
             model=model,
diff --git a/litellm/llms/bedrock/vector_stores/transformation.py b/litellm/llms/bedrock/vector_stores/transformation.py
index 2d72db0cdba..bad17a2181d 100644
--- a/litellm/llms/bedrock/vector_stores/transformation.py
+++ b/litellm/llms/bedrock/vector_stores/transformation.py
@@ -27,6 +27,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.router import Router
 else:
     LiteLLMLoggingObj = Any
 
@@ -196,6 +197,7 @@ class BedrockVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         if isinstance(query, list):
             query = " ".join(query)
diff --git a/litellm/llms/custom_httpx/llm_http_handler.py b/litellm/llms/custom_httpx/llm_http_handler.py
index 834f7d564a2..6f42d42de00 100644
--- a/litellm/llms/custom_httpx/llm_http_handler.py
+++ b/litellm/llms/custom_httpx/llm_http_handler.py
@@ -178,6 +178,7 @@ if TYPE_CHECKING:
         AnthropicMessagesStreamingResponse,
     )
     from litellm.llms.base_llm.passthrough.transformation import BasePassthroughConfig
+    from litellm.router import Router
     from litellm.types.llms.openai_evals import (
         CancelEvalResponse,
         CancelRunResponse,
@@ -2923,7 +2924,7 @@ class BaseLLMHTTPHandler:
         final_response: Final = await self._call_agentic_completion_hooks(
             response=initial_response,
             model=model,
-            messages=(input if isinstance(input, list) else [{"role": "user", "content": input}]),
+            messages=(input if isinstance(input, list) else [{"role": "user", "content": input}]),  # pyright: ignore[reportArgumentType]  # pre-existing mismatch surfaced by the Router import; the hook accepts response input items at runtime
             anthropic_messages_provider_config=responses_api_provider_config,
             anthropic_messages_optional_request_params=response_api_optional_request_params,
             logging_obj=logging_obj,
@@ -5415,7 +5416,7 @@ class BaseLLMHTTPHandler:
         try:
             response: ResponsesAPIResponse | BaseResponsesAPIStreamingIterator = await litellm.aresponses(
                 model=patch.model or model,
-                input=patch.messages,
+                input=patch.messages,  # pyright: ignore[reportArgumentType]  # pre-existing mismatch surfaced by the Router import; patch messages are valid response input at runtime
                 **optional_params,
                 **kwargs_for_followup,
             )
@@ -9688,6 +9689,7 @@ class BaseLLMHTTPHandler:
         timeout: float | httpx.Timeout | None = None,
         client: HTTPHandler | AsyncHTTPHandler | None = None,
         _is_async: bool = False,
+        router: "Router | None" = None,
     ) -> VectorStoreSearchResponse:
         if isinstance(vector_store_provider_config, BaseDirectVectorStoreConfig):
             self._pre_call_direct_vector_store_search(
@@ -9738,6 +9740,7 @@ class BaseLLMHTTPHandler:
                 litellm_logging_obj=logging_obj,
                 litellm_params=dict(litellm_params),
                 extra_body=extra_body,
+                router=router,
             )
         else:
             (
@@ -9751,6 +9754,7 @@ class BaseLLMHTTPHandler:
                 litellm_logging_obj=logging_obj,
                 litellm_params=dict(litellm_params),
                 extra_body=extra_body,
+                router=router,
             )
         all_optional_params: Final[dict[str, object]] = dict(litellm_params)
         all_optional_params.update(vector_store_search_optional_params or {})
@@ -9802,6 +9806,7 @@ class BaseLLMHTTPHandler:
         timeout: float | httpx.Timeout | None = None,
         client: HTTPHandler | AsyncHTTPHandler | None = None,
         _is_async: bool = False,
+        router: "Router | None" = None,
     ) -> VectorStoreSearchResponse | Coroutine[object, object, VectorStoreSearchResponse]:
         if _is_async:
             return self.async_vector_store_search_handler(
@@ -9816,6 +9821,7 @@ class BaseLLMHTTPHandler:
                 extra_body=extra_body,
                 timeout=timeout,
                 client=client,
+                router=router,
             )
 
         if isinstance(vector_store_provider_config, BaseDirectVectorStoreConfig):
@@ -9862,6 +9868,7 @@ class BaseLLMHTTPHandler:
             litellm_logging_obj=logging_obj,
             litellm_params=dict(litellm_params),
             extra_body=extra_body,
+            router=router,
         )
 
         all_optional_params: Final[dict[str, object]] = dict(litellm_params)
diff --git a/litellm/llms/gemini/vector_stores/transformation.py b/litellm/llms/gemini/vector_stores/transformation.py
index f6525a449b6..82586b1f638 100644
--- a/litellm/llms/gemini/vector_stores/transformation.py
+++ b/litellm/llms/gemini/vector_stores/transformation.py
@@ -33,6 +33,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.router import Router
 else:
     LiteLLMLoggingObj = Any
 
@@ -168,6 +169,7 @@ class GeminiVectorStoreConfig(BaseVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: Mapping[str, object] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         """
         Transform search request to Gemini's generateContent format.
diff --git a/litellm/llms/langflow/a2a.py b/litellm/llms/langflow/a2a.py
index cae750d586e..060dc0a4d05 100644
--- a/litellm/llms/langflow/a2a.py
+++ b/litellm/llms/langflow/a2a.py
@@ -1,28 +1,9 @@
-import hashlib
 from typing import Any, Final
 
-
-def get_session_id_from_a2a_params(params: dict[str, Any]) -> str | None:
-    message: Final = params.get("message", {})
-    if isinstance(message, dict):
-        return message.get("contextId")
-    return getattr(message, "contextId", None)
-
-
-def scope_session_to_principal(session_id: str, principal: str | None) -> str:
-    """
-    Bind a client-supplied A2A contextId to the authenticated principal.
-
-    Without this, two distinct keys authorized for the same LangFlow agent could
-    set the same contextId and read/append to each other's LangFlow memory. The
-    principal is hashed (it is already a hashed token) so the raw value is never
-    sent to the LangFlow backend, while the original contextId is kept as a
-    suffix for operator-side correlation.
-    """
-    if not principal:
-        return session_id
-    principal_prefix: Final = hashlib.sha256(principal.encode("utf-8")).hexdigest()[:16]
-    return f"{principal_prefix}-{session_id}"
+from litellm.a2a_protocol.utils import (
+    get_session_id_from_a2a_params,
+    scope_session_to_principal,
+)
 
 
 def merge_a2a_session_into_litellm_params(
diff --git a/litellm/llms/milvus/vector_stores/transformation.py b/litellm/llms/milvus/vector_stores/transformation.py
index 34f0cd854c4..c3581abfbcc 100644
--- a/litellm/llms/milvus/vector_stores/transformation.py
+++ b/litellm/llms/milvus/vector_stores/transformation.py
@@ -19,6 +19,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     LiteLLMLoggingObj = _LiteLLMLoggingObj
 else:
@@ -123,6 +124,7 @@ class MilvusVectorStoreConfig(BaseVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict[str, Any]]:
         """
         Transform search request for Azure AI Search API
diff --git a/litellm/llms/openai/vector_stores/transformation.py b/litellm/llms/openai/vector_stores/transformation.py
index f6c093f2e2a..4e925494039 100644
--- a/litellm/llms/openai/vector_stores/transformation.py
+++ b/litellm/llms/openai/vector_stores/transformation.py
@@ -21,6 +21,7 @@ from litellm.utils import add_openai_metadata
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     LiteLLMLoggingObj = _LiteLLMLoggingObj
 else:
@@ -99,6 +100,7 @@ class OpenAIVectorStoreConfig(BaseVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         encoded_vector_store_id: Final = encode_url_path_segment(vector_store_id, field_name="vector_store_id")
         url: Final = f"{api_base}/{encoded_vector_store_id}/search"
diff --git a/litellm/llms/pg_vector/vector_stores/transformation.py b/litellm/llms/pg_vector/vector_stores/transformation.py
index e4b06c36bf4..9de1f589ae4 100644
--- a/litellm/llms/pg_vector/vector_stores/transformation.py
+++ b/litellm/llms/pg_vector/vector_stores/transformation.py
@@ -8,6 +8,7 @@ from litellm.types.vector_stores import VectorStoreSearchOptionalRequestParams
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.router import Router
 else:
     LiteLLMLoggingObj = Any
 
@@ -80,6 +81,7 @@ class PGVectorStoreConfig(OpenAIVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         encoded_vector_store_id: Final = encode_url_path_segment(vector_store_id, field_name="vector_store_id")
         url: Final = f"{api_base}/{encoded_vector_store_id}/search"
diff --git a/litellm/llms/ragflow/vector_stores/transformation.py b/litellm/llms/ragflow/vector_stores/transformation.py
index 282cb7a92a7..ffa6c9e1076 100644
--- a/litellm/llms/ragflow/vector_stores/transformation.py
+++ b/litellm/llms/ragflow/vector_stores/transformation.py
@@ -17,6 +17,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.router import Router
 else:
     LiteLLMLoggingObj = Any
 
@@ -92,6 +93,7 @@ class RAGFlowVectorStoreConfig(BaseVectorStoreConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         """RAGFlow vector stores are management-only, search is not supported."""
         raise NotImplementedError("RAGFlow vector stores support dataset management only, not search/retrieval")
diff --git a/litellm/llms/s3_vectors/vector_stores/transformation.py b/litellm/llms/s3_vectors/vector_stores/transformation.py
index 5be35ae4148..733358381fe 100644
--- a/litellm/llms/s3_vectors/vector_stores/transformation.py
+++ b/litellm/llms/s3_vectors/vector_stores/transformation.py
@@ -1,8 +1,8 @@
-import re
 from typing import TYPE_CHECKING, Any, Final
 
 import httpx
 
+from litellm.caching._embedding_router import resolve_embedding_router
 from litellm.llms.base_llm.vector_store.transformation import BaseVectorStoreConfig
 from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM
 from litellm.types.router import GenericLiteLLMParams
@@ -18,6 +18,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
+    from litellm.router import Router
 else:
     LiteLLMLoggingObj = Any
 
@@ -58,13 +59,20 @@ class S3VectorsVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
         return headers
 
     def get_complete_url(self, api_base: str | None, litellm_params: dict) -> str:
-        aws_region_name: Final = litellm_params.get("aws_region_name")
-        if not aws_region_name:
-            raise ValueError("aws_region_name is required for S3 Vectors")
-        if not re.match(r"^[a-z][a-z0-9-]*$", aws_region_name):
-            raise ValueError("Invalid aws_region_name format")
+        # Resolve region the same way the ingestion path does:
+        # dynamic param -> AWS_REGION_NAME -> AWS_REGION -> default (us-west-2)
+        aws_region_name: Final = self.get_aws_region_name_for_non_llm_api_calls(litellm_params.get("aws_region_name"))
         return f"https://s3vectors.{aws_region_name}.api.aws"
 
+    def _resolve_query_embedding_router(self, embedding_model: str, router: "Router | None") -> "Router | None":
+        """Return the router iff it serves ``embedding_model`` as a deployment."""
+        if router is None:
+            return None
+        model_list: Final = [
+            dict(m) for m in (router.get_model_list() or ())
+        ]  # mutable-ok: resolve_embedding_router requires list[dict]
+        return resolve_embedding_router(embedding_model=embedding_model, llm_router=router, llm_model_list=model_list)
+
     def transform_search_vector_store_request(
         self,
         vector_store_id: str,
@@ -74,6 +82,7 @@ class S3VectorsVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         """Sync version - generates embedding synchronously."""
         # For S3 Vectors, vector_store_id should be in format: bucket_name:index_name
@@ -99,10 +108,16 @@ class S3VectorsVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
 
         # Generate embedding for the query
         embedding_model: Final = litellm_params.get("embedding_model", "text-embedding-3-small")
+        embedding_router: Final = self._resolve_query_embedding_router(embedding_model=embedding_model, router=router)
 
         import litellm as litellm_module
 
-        embedding_response: Final = litellm_module.embedding(model=embedding_model, input=[query])
+        embedding_input: Final = [query]  # mutable-ok: the embedding API takes list input
+        embedding_response: Final = (
+            embedding_router.embedding(model=embedding_model, input=embedding_input)
+            if embedding_router is not None
+            else litellm_module.embedding(model=embedding_model, input=embedding_input)
+        )
         query_embedding: Final = embedding_response.data[0]["embedding"]
 
         url: Final = f"{api_base}/QueryVectors"
@@ -128,6 +143,7 @@ class S3VectorsVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: dict[str, Any] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict]:
         """Async version - generates embedding asynchronously."""
         # For S3 Vectors, vector_store_id should be in format: bucket_name:index_name
@@ -153,10 +169,16 @@ class S3VectorsVectorStoreConfig(BaseVectorStoreConfig, BaseAWSLLM):
 
         # Generate embedding for the query asynchronously
         embedding_model: Final = litellm_params.get("embedding_model", "text-embedding-3-small")
+        embedding_router: Final = self._resolve_query_embedding_router(embedding_model=embedding_model, router=router)
 
         import litellm as litellm_module
 
-        embedding_response: Final = await litellm_module.aembedding(model=embedding_model, input=[query])
+        embedding_input: Final = [query]  # mutable-ok: the embedding API takes list input
+        embedding_response: Final = (
+            await embedding_router.aembedding(model=embedding_model, input=embedding_input)
+            if embedding_router is not None
+            else await litellm_module.aembedding(model=embedding_model, input=embedding_input)
+        )
         query_embedding: Final = embedding_response.data[0]["embedding"]
 
         url: Final = f"{api_base}/QueryVectors"
diff --git a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
index d8b1e7ba17c..69fe5678de9 100644
--- a/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
+++ b/litellm/llms/vertex_ai/gemini/vertex_and_google_ai_studio_gemini.py
@@ -949,7 +949,7 @@ class VertexGeminiConfig(VertexAIBaseConfig, BaseConfig):
         # For Gemini 3+ models, use thinkingLevel instead of thinkingBudget
         if model and VertexGeminiConfig._is_gemini_3_or_newer(model):
             if thinking_enabled:
-                if thinking_budget is None or thinking_budget == 0:
+                if thinking_budget == 0:
                     params["includeThoughts"] = False
                 else:
                     params["includeThoughts"] = True
diff --git a/litellm/llms/vertex_ai/ocr/deepseek_transformation.py b/litellm/llms/vertex_ai/ocr/deepseek_transformation.py
index 2603552152d..b57a87c3325 100644
--- a/litellm/llms/vertex_ai/ocr/deepseek_transformation.py
+++ b/litellm/llms/vertex_ai/ocr/deepseek_transformation.py
@@ -177,8 +177,9 @@ class VertexAIDeepSeekOCRConfig(BaseOCRConfig):
             content_item = {"type": "image_url", "image_url": document_url}
 
         # Build DeepSeek OCR request
+        provider_model: Final = model if model.startswith("deepseek-ai/") else f"deepseek-ai/{model}"
         data: Final = {
-            "model": "deepseek-ai/" + model,
+            "model": provider_model,
             "messages": [{"role": "user", "content": [content_item]}],
         }
 
diff --git a/litellm/llms/vertex_ai/vector_stores/rag_api/transformation.py b/litellm/llms/vertex_ai/vector_stores/rag_api/transformation.py
index 5c250fc1a7e..36b57e7c995 100644
--- a/litellm/llms/vertex_ai/vector_stores/rag_api/transformation.py
+++ b/litellm/llms/vertex_ai/vector_stores/rag_api/transformation.py
@@ -21,6 +21,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     LiteLLMLoggingObj = _LiteLLMLoggingObj
 else:
@@ -161,6 +162,7 @@ class VertexVectorStoreConfig(BaseVectorStoreConfig, VertexBase):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: Mapping[str, object] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict[str, object]]:
         """
         Transform search request for Vertex AI RAG API
diff --git a/litellm/llms/vertex_ai/vector_stores/search_api/transformation.py b/litellm/llms/vertex_ai/vector_stores/search_api/transformation.py
index 0bcf16ee06f..f0812e3ed9f 100644
--- a/litellm/llms/vertex_ai/vector_stores/search_api/transformation.py
+++ b/litellm/llms/vertex_ai/vector_stores/search_api/transformation.py
@@ -25,6 +25,7 @@ from litellm.types.vector_stores import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
+    from litellm.router import Router
 
     LiteLLMLoggingObj = _LiteLLMLoggingObj
 else:
@@ -245,6 +246,7 @@ class VertexSearchAPIVectorStoreConfig(BaseVectorStoreConfig, VertexBase):
         litellm_logging_obj: LiteLLMLoggingObj,
         litellm_params: dict,
         extra_body: Mapping[str, object] | None = None,
+        router: "Router | None" = None,
     ) -> tuple[str, dict[str, object]]:
         """
         Transform a search request for the Vertex AI Search (Discovery Engine) API.
diff --git a/litellm/main.py b/litellm/main.py
index c4c5bbefc4f..01c106adc7c 100644
--- a/litellm/main.py
+++ b/litellm/main.py
@@ -8637,6 +8637,16 @@ def _set_stream_builder_response_cost(response: ModelResponse, logging_obj: Opti
     hidden_params["response_cost"] = response_cost
 
 
+def _stamp_streaming_usage_cost(usage: Usage, response: ModelResponse, logging_obj: Optional["Logging"]) -> None:
+    if logging_obj is None:
+        return
+    if isinstance(getattr(usage, "cost", None), (int, float)):
+        return
+    computed_cost: Final = logging_obj._response_cost_calculator(result=response)
+    if isinstance(computed_cost, (int, float)) and computed_cost > 0:
+        setattr(usage, "cost", computed_cost)
+
+
 def stream_chunk_builder(
     chunks: list,
     messages: list | None = None,
@@ -8731,12 +8741,7 @@ def stream_chunk_builder(
                     )
                     break
 
-            if litellm.include_cost_in_streaming_usage and logging_obj is not None:
-                setattr(
-                    usage,
-                    "cost",
-                    logging_obj._response_cost_calculator(result=response),
-                )
+            _stamp_streaming_usage_cost(usage, response, logging_obj)
             _set_stream_builder_response_cost(response, logging_obj)
 
             processor.apply_provider_assembled_streaming_metadata(response, chunks, logging_obj)
@@ -8915,10 +8920,7 @@ def stream_chunk_builder(
                 )
                 break
 
-        # Add cost to usage object if include_cost_in_streaming_usage is True
-        if litellm.include_cost_in_streaming_usage and logging_obj is not None:
-            setattr(usage, "cost", logging_obj._response_cost_calculator(result=response))
-
+        _stamp_streaming_usage_cost(usage, response, logging_obj)
         _set_stream_builder_response_cost(response, logging_obj)
 
         processor.apply_provider_assembled_streaming_metadata(response, chunks, logging_obj)
diff --git a/litellm/model_prices_and_context_window_backup.json b/litellm/model_prices_and_context_window_backup.json
index a3cfb300ea6..2846d12db6e 100644
--- a/litellm/model_prices_and_context_window_backup.json
+++ b/litellm/model_prices_and_context_window_backup.json
@@ -9643,7 +9643,8 @@
         "supported_endpoints": [
             "/v1/images/generations",
             "/v1/images/edits"
-        ]
+        ],
+        "deprecation_date": "2026-10-01"
     },
     "azure_ai/MAI-Image-2.5-Flash": {
         "input_cost_per_image_token": 1.75e-06,
@@ -9656,7 +9657,8 @@
         "supported_endpoints": [
             "/v1/images/generations",
             "/v1/images/edits"
-        ]
+        ],
+        "deprecation_date": "2026-10-01"
     },
     "azure_ai/MAI-Image-2e": {
         "deprecation_date": "2026-08-15",
@@ -10155,7 +10157,9 @@
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_reasoning": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "cache_read_input_token_cost": 1.45e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/deepseek-v4-flash": {
         "deprecation_date": "2028-02-20",
@@ -10169,18 +10173,20 @@
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_reasoning": true,
-        "supports_tool_choice": true
-    },
-    "azure_ai/deepseek-v4-flash-0731": {
+        "supports_tool_choice": true,
         "cache_read_input_token_cost": 2.8e-08,
+        "supports_prompt_caching": true
+    },
+    "azure_ai/DeepSeek-V4-Flash-0731": {
+        "cache_read_input_token_cost": 1.4e-08,
         "deprecation_date": "2026-12-03",
-        "input_cost_per_token": 1.9e-07,
+        "input_cost_per_token": 4.4e-07,
         "litellm_provider": "azure_ai",
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000,
         "max_tokens": 128000,
         "mode": "chat",
-        "output_cost_per_token": 5.1e-07,
+        "output_cost_per_token": 1.32e-06,
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_prompt_caching": true,
@@ -10400,11 +10406,13 @@
         "max_tokens": 262144,
         "mode": "chat",
         "output_cost_per_token": 3e-06,
-        "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/kimi-k2-5-now-in-microsoft-foundry/4492321",
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
         "supports_function_calling": true,
         "supports_tool_choice": true,
         "supports_video_input": true,
-        "supports_vision": true
+        "supports_vision": true,
+        "cache_read_input_token_cost": 1e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/kimi-k2.6": {
         "deprecation_date": "2027-04-16",
@@ -10415,7 +10423,7 @@
         "max_tokens": 262144,
         "mode": "chat",
         "output_cost_per_token": 4e-06,
-        "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-kimi-k2-6-in-microsoft-foundry/4513125",
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
         "supported_modalities": [
             "text",
             "image"
@@ -10426,7 +10434,9 @@
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_tool_choice": true,
-        "supports_vision": true
+        "supports_vision": true,
+        "cache_read_input_token_cost": 1.6e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/ministral-3b": {
         "input_cost_per_token": 4e-08,
@@ -12110,7 +12120,7 @@
         "max_output_tokens": 2048,
         "max_tokens": 2048,
         "mode": "chat",
-        "output_cost_per_token": 2.65e-06,
+        "output_cost_per_token": 6e-07,
         "supports_pdf_input": true
     },
     "bedrock/us-west-1/meta.llama3-70b-instruct-v1:0": {
@@ -23514,6 +23524,63 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "vertex_ai/gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "vertex_ai",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "regional_endpoint_uplift_multiplier": 1.1,
+        "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "vertex_ai/gemini-3.1-pro-preview": {
         "prompt_cache_min_tokens": 4096,
         "cache_read_input_token_cost": 2e-07,
@@ -25351,6 +25418,65 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "gemini/gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "gemini",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "rpm": 2000,
+        "source": "https://ai.google.dev/gemini-api/docs/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_output": false,
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "tpm": 800000,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "gemini/gemini-omni-flash-preview": {
         "input_cost_per_audio_token": 1.5e-06,
         "input_cost_per_token": 1.5e-06,
@@ -25759,6 +25885,63 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "vertex_ai-language-models",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "source": "https://ai.google.dev/gemini-api/docs/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_output": false,
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "gemini/gemini-2.5-pro-preview-tts": {
         "cache_read_input_token_cost": 1.25e-07,
         "input_cost_per_audio_token": 7e-07,
@@ -29098,16 +29281,19 @@
         "cache_creation_input_token_cost": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 1e-05,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
         "cache_creation_input_token_cost_flex": 2.5e-06,
         "cache_creation_input_token_cost_priority": 1e-05,
         "cache_read_input_token_cost": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens": 8e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 4e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
         "cache_read_input_token_cost_flex": 2e-07,
         "cache_read_input_token_cost_priority": 8e-07,
         "input_cost_per_token": 4e-06,
         "input_cost_per_token_above_272k_tokens": 8e-06,
         "input_cost_per_token_above_272k_tokens_flex": 4e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
         "input_cost_per_token_batches": 2e-06,
         "input_cost_per_token_flex": 2e-06,
         "input_cost_per_token_priority": 8e-06,
@@ -29119,6 +29305,7 @@
         "output_cost_per_token": 2e-05,
         "output_cost_per_token_above_272k_tokens": 3e-05,
         "output_cost_per_token_above_272k_tokens_flex": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
         "output_cost_per_token_batches": 1e-05,
         "output_cost_per_token_flex": 1e-05,
         "output_cost_per_token_priority": 4e-05,
@@ -29161,16 +29348,19 @@
         "cache_creation_input_token_cost": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 1e-05,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
         "cache_creation_input_token_cost_flex": 2.5e-06,
         "cache_creation_input_token_cost_priority": 1e-05,
         "cache_read_input_token_cost": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens": 8e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 4e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
         "cache_read_input_token_cost_flex": 2e-07,
         "cache_read_input_token_cost_priority": 8e-07,
         "input_cost_per_token": 4e-06,
         "input_cost_per_token_above_272k_tokens": 8e-06,
         "input_cost_per_token_above_272k_tokens_flex": 4e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
         "input_cost_per_token_batches": 2e-06,
         "input_cost_per_token_flex": 2e-06,
         "input_cost_per_token_priority": 8e-06,
@@ -29182,6 +29372,7 @@
         "output_cost_per_token": 2e-05,
         "output_cost_per_token_above_272k_tokens": 3e-05,
         "output_cost_per_token_above_272k_tokens_flex": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
         "output_cost_per_token_batches": 1e-05,
         "output_cost_per_token_flex": 1e-05,
         "output_cost_per_token_priority": 4e-05,
@@ -29225,16 +29416,19 @@
         "cache_creation_input_token_cost": 2.5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 2.5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-05,
         "cache_creation_input_token_cost_flex": 1.25e-06,
         "cache_creation_input_token_cost_priority": 5e-06,
         "cache_read_input_token_cost": 2e-07,
         "cache_read_input_token_cost_above_272k_tokens": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 2e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-07,
         "cache_read_input_token_cost_flex": 1e-07,
         "cache_read_input_token_cost_priority": 4e-07,
         "input_cost_per_token": 2e-06,
         "input_cost_per_token_above_272k_tokens": 4e-06,
         "input_cost_per_token_above_272k_tokens_flex": 2e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 8e-06,
         "input_cost_per_token_batches": 1e-06,
         "input_cost_per_token_flex": 1e-06,
         "input_cost_per_token_priority": 4e-06,
@@ -29246,6 +29440,7 @@
         "output_cost_per_token": 1.2e-05,
         "output_cost_per_token_above_272k_tokens": 1.8e-05,
         "output_cost_per_token_above_272k_tokens_flex": 9e-06,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-05,
         "output_cost_per_token_batches": 6e-06,
         "output_cost_per_token_flex": 6e-06,
         "output_cost_per_token_priority": 2.4e-05,
@@ -29288,16 +29483,19 @@
         "cache_creation_input_token_cost": 2.5e-07,
         "cache_creation_input_token_cost_above_272k_tokens": 5e-07,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 2.5e-07,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-06,
         "cache_creation_input_token_cost_flex": 1.25e-07,
         "cache_creation_input_token_cost_priority": 5e-07,
         "cache_read_input_token_cost": 2e-08,
         "cache_read_input_token_cost_above_272k_tokens": 4e-08,
         "cache_read_input_token_cost_above_272k_tokens_flex": 2e-08,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08,
         "cache_read_input_token_cost_flex": 1e-08,
         "cache_read_input_token_cost_priority": 4e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_above_272k_tokens": 4e-07,
         "input_cost_per_token_above_272k_tokens_flex": 2e-07,
+        "input_cost_per_token_above_272k_tokens_priority": 8e-07,
         "input_cost_per_token_batches": 1e-07,
         "input_cost_per_token_flex": 1e-07,
         "input_cost_per_token_priority": 4e-07,
@@ -29309,6 +29507,7 @@
         "output_cost_per_token": 1.2e-06,
         "output_cost_per_token_above_272k_tokens": 1.8e-06,
         "output_cost_per_token_above_272k_tokens_flex": 9e-07,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-06,
         "output_cost_per_token_batches": 6e-07,
         "output_cost_per_token_flex": 6e-07,
         "output_cost_per_token_priority": 2.4e-06,
@@ -29548,7 +29747,10 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": false
+        "supports_minimal_reasoning_effort": false,
+        "input_cost_per_token_above_272k_tokens_flex": 5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07
     },
     "gpt-5.5-2026-04-23": {
         "cache_read_input_token_cost": 5e-07,
@@ -29602,7 +29804,10 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": false
+        "supports_minimal_reasoning_effort": false,
+        "input_cost_per_token_above_272k_tokens_flex": 5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07
     },
     "gpt-5.5-pro": {
         "cache_read_input_token_cost": 3e-06,
@@ -29751,7 +29956,10 @@
         "supports_none_reasoning_effort": true,
         "default_reasoning_effort": "none",
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07
     },
     "gpt-5.4-2026-03-05": {
         "cache_read_input_token_cost": 2.5e-07,
@@ -29800,7 +30008,10 @@
         "supports_none_reasoning_effort": true,
         "default_reasoning_effort": "none",
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07
     },
     "gpt-5.4-pro": {
         "cache_read_input_token_cost": 3e-06,
@@ -29849,7 +30060,9 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": false,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 3e-05,
+        "output_cost_per_token_above_272k_tokens_flex": 0.000135
     },
     "gpt-5.4-pro-2026-03-05": {
         "cache_read_input_token_cost": 3e-06,
@@ -29898,7 +30111,9 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": false,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 3e-05,
+        "output_cost_per_token_above_272k_tokens_flex": 0.000135
     },
     "gpt-5.4-mini": {
         "cache_read_input_token_cost": 7.5e-08,
@@ -30834,17 +31049,18 @@
     },
     "gpt-realtime-2": {
         "cache_creation_input_audio_token_cost": 4e-07,
+        "cache_read_input_audio_token_cost": 4e-07,
         "cache_read_input_token_cost": 4e-07,
         "input_cost_per_audio_token": 3.2e-05,
         "input_cost_per_image": 5e-06,
         "input_cost_per_token": 4e-06,
         "litellm_provider": "openai",
-        "max_input_tokens": 32000,
-        "max_output_tokens": 4096,
-        "max_tokens": 4096,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
         "mode": "realtime",
         "output_cost_per_audio_token": 6.4e-05,
-        "output_cost_per_token": 1.6e-05,
+        "output_cost_per_token": 2.4e-05,
         "supported_endpoints": [
             "/v1/realtime"
         ],
@@ -30908,8 +31124,8 @@
         "input_cost_per_token": 6e-07,
         "litellm_provider": "openai",
         "max_input_tokens": 128000,
-        "max_output_tokens": 4096,
-        "max_tokens": 4096,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
         "mode": "realtime",
         "output_cost_per_audio_token": 2e-05,
         "output_cost_per_token": 2.4e-06,
@@ -30941,7 +31157,7 @@
         "input_cost_per_audio_token": 1e-05,
         "input_cost_per_token": 6e-07,
         "litellm_provider": "openai",
-        "max_input_tokens": 128000,
+        "max_input_tokens": 32000,
         "max_output_tokens": 4096,
         "max_tokens": 4096,
         "mode": "realtime",
@@ -33705,19 +33921,21 @@
         "source": "https://mistral.ai/pricing#api-pricing"
     },
     "mistral/magistral-medium-latest": {
-        "input_cost_per_token": 2e-06,
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.5e-06,
         "litellm_provider": "mistral",
-        "max_input_tokens": 40000,
-        "max_output_tokens": 40000,
-        "max_tokens": 40000,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 5e-06,
-        "source": "https://mistral.ai/news/magistral",
+        "output_cost_per_token": 7.5e-06,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04",
         "supports_assistant_prefill": true,
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/magistral-small-2506": {
         "deprecation_date": "2025-11-30",
@@ -33736,19 +33954,21 @@
         "supports_tool_choice": true
     },
     "mistral/magistral-small-latest": {
-        "input_cost_per_token": 5e-07,
+        "cache_read_input_token_cost": 1.5e-08,
+        "input_cost_per_token": 1.5e-07,
         "litellm_provider": "mistral",
-        "max_input_tokens": 40000,
-        "max_output_tokens": 40000,
-        "max_tokens": 40000,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 1.5e-06,
-        "source": "https://mistral.ai/pricing#api-pricing",
+        "output_cost_per_token": 6e-07,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03",
         "supports_assistant_prefill": true,
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/magistral-small-1-2-2509": {
         "deprecation_date": "2026-07-31",
@@ -33880,16 +34100,21 @@
         "supports_vision": true
     },
     "mistral/mistral-medium": {
-        "input_cost_per_token": 2.7e-06,
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.5e-06,
         "litellm_provider": "mistral",
-        "max_input_tokens": 32000,
-        "max_output_tokens": 8191,
-        "max_tokens": 8191,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 8.1e-06,
+        "output_cost_per_token": 7.5e-06,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04",
         "supports_assistant_prefill": true,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/mistral-medium-2312": {
         "deprecation_date": "2025-06-16",
@@ -41340,13 +41565,13 @@
         "source": "https://docs.together.ai/docs/serverless-models"
     },
     "together_ai/Qwen/Qwen3.8-2.4T-A95B": {
-        "cache_read_input_token_cost": 5e-07,
-        "input_cost_per_token": 2.5e-06,
+        "cache_read_input_token_cost": 2.5e-07,
+        "input_cost_per_token": 2e-06,
         "litellm_provider": "together_ai",
         "max_input_tokens": 1010000,
         "max_tokens": 1010000,
         "mode": "chat",
-        "output_cost_per_token": 6.25e-06,
+        "output_cost_per_token": 6e-06,
         "source": "https://docs.together.ai/docs/serverless-models",
         "supports_prompt_caching": true
     },
@@ -41956,6 +42181,70 @@
         "supports_parallel_tool_use_config": true,
         "prompt_cache_min_tokens": 1024
     },
+    "us-gov.anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock_converse",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "us-gov.anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock_converse",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
     "au.anthropic.claude-haiku-4-5-20251001-v1:0": {
         "cache_creation_input_token_cost": 1.375e-06,
         "cache_creation_input_token_cost_above_1hr": 2.2e-06,
@@ -45849,6 +46138,26 @@
         "mode": "rerank",
         "output_cost_per_token": 0.0
     },
+    "voyage/rerank-3": {
+        "input_cost_per_token": 5e-08,
+        "litellm_provider": "voyage",
+        "max_input_tokens": 32000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
+        "mode": "rerank",
+        "output_cost_per_token": 0.0,
+        "source": "https://docs.voyageai.com/docs/pricing"
+    },
+    "voyage/rerank-3-lite": {
+        "input_cost_per_token": 2e-08,
+        "litellm_provider": "voyage",
+        "max_input_tokens": 32000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
+        "mode": "rerank",
+        "output_cost_per_token": 0.0,
+        "source": "https://docs.voyageai.com/docs/pricing"
+    },
     "voyage/voyage-2": {
         "input_cost_per_token": 1e-07,
         "litellm_provider": "voyage",
@@ -47100,6 +47409,27 @@
         "supports_vision": true,
         "supports_web_search": true
     },
+    "xai/grok-build-latest": {
+        "cache_read_input_token_cost": 3e-07,
+        "cache_read_input_token_cost_above_200k_tokens": 6e-07,
+        "input_cost_per_token": 2e-06,
+        "input_cost_per_token_above_200k_tokens": 4e-06,
+        "litellm_provider": "xai",
+        "max_input_tokens": 500000,
+        "max_output_tokens": 500000,
+        "max_tokens": 500000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-06,
+        "output_cost_per_token_above_200k_tokens": 1.2e-05,
+        "source": "https://docs.x.ai/developers/models",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true
+    },
     "xai/grok-4.6": {
         "cache_read_input_token_cost": 5e-07,
         "cache_read_input_token_cost_above_200k_tokens": 1e-06,
@@ -57420,6 +57750,34 @@
         "supports_tool_choice": true,
         "supports_vision": false
     },
+    "fireworks_ai/accounts/fireworks/models/glm-5p3-flash": {
+        "cache_read_input_token_cost": 3e-08,
+        "input_cost_per_token": 1.5e-07,
+        "litellm_provider": "fireworks_ai",
+        "max_input_tokens": 1048576,
+        "max_tokens": 1048576,
+        "mode": "chat",
+        "output_cost_per_token": 5e-07,
+        "source": "https://docs.fireworks.ai/serverless/pricing",
+        "supports_function_calling": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "fireworks_ai/accounts/fireworks/models/inkling": {
+        "cache_read_input_token_cost": 1.7e-07,
+        "input_cost_per_token": 1e-06,
+        "litellm_provider": "fireworks_ai",
+        "max_input_tokens": 1048576,
+        "max_tokens": 1048576,
+        "mode": "chat",
+        "output_cost_per_token": 4.05e-06,
+        "source": "https://fireworks.ai/models/fireworks/inkling",
+        "supports_function_calling": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
     "fireworks_ai/accounts/fireworks/models/qwen3-embedding-8b": {
         "input_cost_per_token": 1e-07,
         "output_cost_per_token": 0.0,
@@ -57479,5 +57837,542 @@
         "supported_endpoints": [
             "/v1/audio/transcriptions"
         ]
+    },
+    "scaleway/glm-5.2": {
+        "input_cost_per_token": 1.8e-06,
+        "litellm_provider": "scaleway",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 16384,
+        "max_tokens": 16384,
+        "mode": "chat",
+        "output_cost_per_token": 5.5e-06,
+        "source": "https://www.scaleway.com/en/pricing/model-as-a-service/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_vision": false
+    },
+    "scaleway/deepseek-v4-flash-0731": {
+        "cache_read_input_token_cost": 8e-08,
+        "input_cost_per_token": 4e-07,
+        "litellm_provider": "scaleway",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 8e-07,
+        "source": "https://www.scaleway.com/en/pricing/model-as-a-service/",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_vision": false
+    },
+    "azure_ai/kimi-k2.7-code": {
+        "cache_read_input_token_cost": 1.9e-07,
+        "deprecation_date": "2026-10-03",
+        "input_cost_per_token": 9.5e-07,
+        "litellm_provider": "azure_ai",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
+        "mode": "chat",
+        "output_cost_per_token": 4e-06,
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-nano-3-30b": {
+        "input_cost_per_token": 7.2e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 2.88e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_native_structured_output": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-nano-12b-v2": {
+        "input_cost_per_token": 2.4e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_system_messages": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-super-3-120b": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 7.8e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/openai.gpt-oss-20b-1:0": {
+        "input_cost_per_token": 8.4e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.6e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/openai.gpt-oss-120b-1:0": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-nano-3-30b": {
+        "input_cost_per_token": 7.2e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 2.88e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_native_structured_output": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-nano-12b-v2": {
+        "input_cost_per_token": 2.4e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_system_messages": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-super-3-120b": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 7.8e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/openai.gpt-oss-20b-1:0": {
+        "input_cost_per_token": 8.4e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.6e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/openai.gpt-oss-120b-1:0": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.6-terra": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 2.64e-06,
+        "input_cost_per_token_above_272k_tokens": 5.28e-06,
+        "cache_creation_input_token_cost": 3.3e-06,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-06,
+        "cache_read_input_token_cost": 2.64e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-07,
+        "output_cost_per_token": 1.584e-05,
+        "output_cost_per_token_above_272k_tokens": 2.376e-05
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.6-luna": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 2.64e-07,
+        "input_cost_per_token_above_272k_tokens": 5.28e-07,
+        "cache_creation_input_token_cost": 3.3e-07,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-07,
+        "cache_read_input_token_cost": 2.64e-08,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-08,
+        "output_cost_per_token": 1.584e-06,
+        "output_cost_per_token_above_272k_tokens": 2.376e-06
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.4": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 3.3e-06,
+        "cache_read_input_token_cost": 3.3e-07,
+        "output_cost_per_token": 1.98e-05
+    },
+    "bedrock_mantle/us-gov-west-1/xai.grok-4.3": {
+        "use_openai_responses_path": true,
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 131072,
+        "max_output_tokens": 16384,
+        "max_tokens": 16384,
+        "mode": "chat",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "input_cost_per_token": 1.5e-06,
+        "output_cost_per_token": 3e-06,
+        "cache_read_input_token_cost": 2.4e-07
+    },
+    "bedrock_mantle/us-gov-east-1/openai.gpt-5.4": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 3.3e-06,
+        "cache_read_input_token_cost": 3.3e-07,
+        "output_cost_per_token": 1.98e-05
+    },
+    "azure/us-gov/gpt-5.1": {
+        "cache_read_input_token_cost": 1.71875e-07,
+        "default_reasoning_effort": "none",
+        "input_cost_per_token": 1.71875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 272000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.375e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text",
+            "image"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_none_reasoning_effort": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "azure/us-gov/o3-mini": {
+        "cache_read_input_token_cost": 7.57e-07,
+        "input_cost_per_token": 1.513e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 200000,
+        "max_output_tokens": 100000,
+        "max_tokens": 100000,
+        "mode": "chat",
+        "output_cost_per_token": 6.05e-06,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": false
+    },
+    "azure/us-gov/text-embedding-3-large": {
+        "input_cost_per_token": 1.63e-07,
+        "litellm_provider": "azure",
+        "max_input_tokens": 8191,
+        "max_tokens": 8191,
+        "mode": "embedding",
+        "output_cost_per_token": 0.0
+    },
+    "azure/us-gov/text-embedding-3-small": {
+        "input_cost_per_token": 2.5e-08,
+        "litellm_provider": "azure",
+        "max_input_tokens": 8191,
+        "max_tokens": 8191,
+        "mode": "embedding",
+        "output_cost_per_token": 0.0
+    },
+    "cloudflare/@cf/openai/whisper": {
+        "input_cost_per_second": 7.5e-06,
+        "litellm_provider": "cloudflare",
+        "mode": "audio_transcription",
+        "output_cost_per_second": 0.0,
+        "source": "https://developers.cloudflare.com/workers-ai/models/whisper/",
+        "supported_endpoints": [
+            "/v1/audio/transcriptions"
+        ]
+    },
+    "cloudflare/@cf/openai/whisper-large-v3-turbo": {
+        "input_cost_per_second": 8.5e-06,
+        "litellm_provider": "cloudflare",
+        "mode": "audio_transcription",
+        "output_cost_per_second": 0.0,
+        "source": "https://developers.cloudflare.com/workers-ai/models/whisper-large-v3-turbo/",
+        "supported_endpoints": [
+            "/v1/audio/transcriptions"
+        ]
     }
 }
diff --git a/litellm/proxy/agent_endpoints/a2a_endpoints.py b/litellm/proxy/agent_endpoints/a2a_endpoints.py
index 31b05320cd3..28882484db4 100644
--- a/litellm/proxy/agent_endpoints/a2a_endpoints.py
+++ b/litellm/proxy/agent_endpoints/a2a_endpoints.py
@@ -1019,4 +1019,6 @@ async def invoke_agent_a2a(
             )
         except Exception:
             pass
+        if isinstance(e, litellm.BadRequestError):
+            return _jsonrpc_error(body.get("id"), -32602, e.message, 400)
         return _jsonrpc_error(body.get("id"), -32603, f"Internal error: {e}", 500)
diff --git a/litellm/proxy/common_utils/model_listing_utils.py b/litellm/proxy/common_utils/model_listing_utils.py
index 9fd24162f7e..213a697b3dd 100644
--- a/litellm/proxy/common_utils/model_listing_utils.py
+++ b/litellm/proxy/common_utils/model_listing_utils.py
@@ -10,13 +10,36 @@ legacy internal names with `general_settings.use_team_public_model_name: false`.
 
 from __future__ import annotations
 
-from collections.abc import Mapping
+from collections.abc import Mapping, Sequence
+from types import MappingProxyType
 from typing import TYPE_CHECKING, Final, cast
 
 if TYPE_CHECKING:
     from litellm.router import Router
 
 
+def configured_display_names(
+    entries: Sequence[tuple[str, str]],
+    llm_router: Router | None,
+) -> Mapping[str, str]:
+    """response_id -> configured `model_info.display_name` for the listing entries
+    that have one.
+
+    Metadata is looked up by each entry's internal lookup id (so team-scoped rows
+    resolve), while the returned map is keyed by the public response id the
+    Anthropic-shaped listing is built from. Entries without a configured name are
+    omitted so the listing falls back to the id itself.
+    """
+    if llm_router is None:
+        return MappingProxyType({})
+    resolved: Final = (
+        (response_id, llm_router.get_configured_display_name(lookup_id)) for response_id, lookup_id in entries
+    )
+    return MappingProxyType(
+        {response_id: display_name for response_id, display_name in resolved if display_name is not None}
+    )
+
+
 class TeamModelNameTranslator:
     """Translates internal team routing keys to their public names for the model
     listing/retrieve responses. Stateless; the live router and general_settings
diff --git a/litellm/proxy/container_endpoints/endpoints.py b/litellm/proxy/container_endpoints/endpoints.py
index 4a088140725..eaa3db336a9 100644
--- a/litellm/proxy/container_endpoints/endpoints.py
+++ b/litellm/proxy/container_endpoints/endpoints.py
@@ -208,7 +208,7 @@ async def list_containers(
 
     # Read query parameters
     query_params: Final = dict(request.query_params)
-    data: Final[dict[str, Any]] = {"query_params": query_params}
+    data: Final[dict[str, Any]] = {"query_params": query_params, "model": query_params.get("model")}
 
     # Extract custom_llm_provider using priority chain
     custom_llm_provider: Final = (
diff --git a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
index d8c8c2f4974..fc881a60f43 100644
--- a/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
+++ b/litellm/proxy/guardrails/guardrail_hooks/headroom/headroom.py
@@ -10,6 +10,7 @@ from typing import TYPE_CHECKING, Any, ClassVar, Final, Literal, TypeGuard
 import httpx
 from fastapi import HTTPException
 from httpx import Response as HttpxResponse
+from pydantic import TypeAdapter
 
 import litellm
 from litellm._logging import verbose_proxy_logger
@@ -52,6 +53,10 @@ BYPASS_HEADER: Final = "x-headroom-bypass"
 HEADROOM_RETRIEVE_TOOL_NAME: Final = "headroom_retrieve"
 _HASH_PATTERN: Final = re.compile(r"hash=([a-f0-9]{24})")
 _HASH_CACHE_TTL_SECONDS: Final = 15 * 60
+# Narrows the base class's bare-dict ``request_data`` at the boundary so its
+# untranslated messages can be read with concrete types (values pass through by
+# reference, so this is a shallow top-level reconstruction).
+_REQUEST_DATA_ADAPTER: Final = TypeAdapter(dict[str, object])
 
 
 def _is_str_object_dict(value: object) -> TypeGuard[dict[str, object]]:  # guard-ok: isinstance narrows correctly; predicate is trivially correct  # fmt: skip
@@ -116,16 +121,119 @@ def _restore_content_shapes(
     return restored
 
 
-def _protected_indices(messages: Sequence[Mapping[str, object]]) -> frozenset[int]:
+def _tool_call_name(tool_call: Mapping[str, object]) -> str | None:
+    function: Final = tool_call.get("function")
+    if not _is_str_object_dict(function):
+        return None
+    name: Final = function.get("name")
+    return name if isinstance(name, str) else None
+
+
+def _is_retrieve_tool_name(name: str | None) -> bool:
+    """Match the retrieve tool whether called directly or via the MCP gateway.
+
+    Server-side the tool is ``headroom_retrieve``; exposed through LiteLLM's MCP
+    gateway a client calls it as ``mcp__<server>__headroom_retrieve``.
+    """
+    return name is not None and (
+        name == HEADROOM_RETRIEVE_TOOL_NAME or name.endswith(f"__{HEADROOM_RETRIEVE_TOOL_NAME}")
+    )
+
+
+def _retrieve_call_ids_in_message(message: Mapping[str, object]) -> frozenset[str]:
+    if message.get("role") != "assistant":
+        return frozenset()
+    tool_calls: Final = message.get("tool_calls")
+    if not _is_object_list(tool_calls):
+        return frozenset()
+    return frozenset(
+        str(tool_call["id"])
+        for tool_call in tool_calls
+        if _is_str_object_dict(tool_call) and tool_call.get("id") and _is_retrieve_tool_name(_tool_call_name(tool_call))
+    )
+
+
+def _anthropic_tool_use_retrieve_id(block: object) -> str | None:
+    if not _is_str_object_dict(block) or block.get("type") != "tool_use":
+        return None
+    name: Final = block.get("name")
+    call_id: Final = block.get("id")
+    if isinstance(name, str) and call_id is not None and _is_retrieve_tool_name(name):
+        return str(call_id)
+    return None
+
+
+def _anthropic_retrieve_ids_in_message(message: Mapping[str, object]) -> frozenset[str]:
+    content: Final = message.get("content")
+    if not _is_object_list(content):
+        return frozenset()
+    return frozenset(call_id for block in content if (call_id := _anthropic_tool_use_retrieve_id(block)) is not None)
+
+
+def _raw_retrieve_call_ids(messages: object) -> frozenset[str]:
+    """Retrieve-tool call ids read from the request's own, untranslated messages.
+
+    The guardrail otherwise scans an OpenAI-translated view where a tool name
+    over 64 chars is truncated to ``{prefix}_{hash}``, which drops the
+    ``__headroom_retrieve`` suffix a long ``mcp__<server>__`` prefix pushes past
+    the limit. Tool-call ids are never truncated, so pairing the tool result to
+    an id read from the original request keeps the match intact. Both wire
+    shapes are handled: OpenAI ``tool_calls`` and Anthropic ``tool_use`` blocks.
+    """
+    if not _is_object_list(messages):
+        return frozenset()
+    return frozenset(
+        call_id
+        for message in messages
+        if _is_str_object_dict(message)
+        for call_id in _retrieve_call_ids_in_message(message) | _anthropic_retrieve_ids_in_message(message)
+    )
+
+
+def _retrieval_result_indices(
+    messages: Sequence[Mapping[str, object]], extra_retrieve_call_ids: frozenset[str] = frozenset()
+) -> frozenset[int]:
+    """Indices of tool-result rows that carry ``headroom_retrieve`` output.
+
+    When the retrieve tool is exposed to a client that runs its own tool loop
+    (the LiteLLM MCP gateway path), the client executes the call and sends the
+    recovered original content back as a tool result on the next turn. That
+    content is exactly what a prior compression stubbed, so compressing it again
+    re-derives the identical content hash: a no-op that strands the model on the
+    marker and loops the agent. Hold those rows back so the expansion survives.
+
+    ``extra_retrieve_call_ids`` carries ids recovered from the untruncated
+    request so the pairing survives tool-name truncation (see
+    ``_raw_retrieve_call_ids``).
+    """
+    retrieve_call_ids: Final = extra_retrieve_call_ids | frozenset(
+        call_id for message in messages for call_id in _retrieve_call_ids_in_message(message)
+    )
+    if not retrieve_call_ids:
+        return frozenset()
+    return frozenset(
+        index
+        for index, message in enumerate(messages)
+        if message.get("role") in ("tool", "function") and str(message.get("tool_call_id")) in retrieve_call_ids
+    )
+
+
+def _protected_indices(
+    messages: Sequence[Mapping[str, object]], extra_retrieve_call_ids: frozenset[str] = frozenset()
+) -> frozenset[int]:
     """Indices headroom must not send to the compression service.
 
     ``get_protected_indices`` is litellm's own compression policy: the system
-    rows, the last user row, the last assistant row. It is expanded over whole
+    rows, the last user row, the last assistant row. Rows carrying just-retrieved
+    ``headroom_retrieve`` output are added so re-compression can't collapse them
+    back to the marker they were expanded from. The union is expanded over whole
     tool exchanges the way ``compress()`` expands it, so a protected assistant
     tool call cannot end up answered by a marker standing in for the result the
     model just asked for.
     """
-    protected: Final = frozenset(get_protected_indices(messages))
+    protected: Final = frozenset(get_protected_indices(messages)) | _retrieval_result_indices(
+        messages, extra_retrieve_call_ids
+    )
     return protected | frozenset(
         index
         for group in group_tool_exchanges(messages)
@@ -634,7 +742,11 @@ class HeadroomGuardrail(CustomGuardrail):
         # /v1/compress grows a field for sending the live turn as the retrieval
         # query without compressing it: query-aware compression reads the newest
         # user message, so it is withheld here at some cost to history ranking.
-        protected_indices: Final = _protected_indices(messages)
+        # request_data is a bare dict on the base signature; narrow it before
+        # reading the untranslated messages so long tool names can be recovered.
+        raw_messages: Final = _REQUEST_DATA_ADAPTER.validate_python(request_data).get("messages")
+        raw_retrieve_call_ids: Final = _raw_retrieve_call_ids(raw_messages)
+        protected_indices: Final = _protected_indices(messages, raw_retrieve_call_ids)
         compressible: Final = [m for i, m in enumerate(messages) if i not in protected_indices]
         if not compressible:
             return inputs
diff --git a/litellm/proxy/hooks/model_max_budget_limiter.py b/litellm/proxy/hooks/model_max_budget_limiter.py
index c5d10b2749b..efaaab277a9 100644
--- a/litellm/proxy/hooks/model_max_budget_limiter.py
+++ b/litellm/proxy/hooks/model_max_budget_limiter.py
@@ -1,6 +1,6 @@
 import json
 import time
-from collections.abc import Iterable, Mapping
+from collections.abc import Iterable, Mapping, Sequence
 from dataclasses import dataclass
 from types import MappingProxyType
 from typing import Final
@@ -199,18 +199,10 @@ async def build_model_max_budget_usage(
         )
         for budget_model, budget_config in budgets
     )
-    batched: Final = await cache.async_batch_get_cache(
-        keys=list(spend_keys)  # mutable-ok: async_batch_get_cache annotates keys as list, so one must exist here
-    )
-    # async_batch_get_cache returns None if it fails internally, and its result is
-    # index-aligned with `keys` otherwise. An unusable result reads as a miss,
-    # which is what a never-written counter already reads as.
-    current_spends: Final = (
-        tuple(batched) if isinstance(batched, list) and len(batched) == len(budgets) else (None,) * len(budgets)
-    )
+    current_spends: Final = await _current_window_spends(cache=cache, spend_keys=spend_keys)
     return {
         budget_model: {
-            "current_spend": round(_as_spend(current_spend), 4),
+            "current_spend": round(current_spend, 4),
             "budget_limit": budget_config.max_budget,
             "time_period": budget_config.budget_duration,
         }
@@ -218,6 +210,22 @@ async def build_model_max_budget_usage(
     }
 
 
+async def _current_window_spends(cache: DualCache, spend_keys: Sequence[str]) -> tuple[float, ...]:
+    """Redis holds the window total across replicas; the in-memory copy is one replica's share."""
+    keys: Final = list(spend_keys)  # mutable-ok: both batch readers annotate their key argument as list
+    redis_cache: Final = cache.redis_cache
+    if redis_cache is not None:
+        shared: Final = await redis_cache.async_batch_get_cache(key_list=keys)
+        return tuple(_as_spend(shared.get(key)) for key in keys)
+    # async_batch_get_cache returns None if it fails internally, and its result is
+    # index-aligned with `keys` otherwise. An unusable result reads as a miss,
+    # which is what a never-written counter already reads as.
+    batched: Final = await cache.async_batch_get_cache(keys=keys)
+    if not isinstance(batched, list) or len(batched) != len(keys):
+        return (0.0,) * len(keys)
+    return tuple(_as_spend(current_spend) for current_spend in batched)
+
+
 def _usable_budget_config(raw_budget_config: object) -> BudgetConfig | None:
     try:
         budget_config: Final = BudgetConfig.model_validate(raw_budget_config)
@@ -404,7 +412,10 @@ class _PROXY_VirtualKeyModelMaxBudgetLimiter(RouterBudgetLimiting):
         return current_spend + _as_spend(await self._cached_spend(legacy_spend_key))
 
     async def _cached_spend(self, spend_key: str) -> float | None:
-        return await self.dual_cache.async_get_cache(key=spend_key)
+        redis_cache: Final = self.dual_cache.redis_cache
+        if redis_cache is None:
+            return await self.dual_cache.async_get_cache(key=spend_key)
+        return await redis_cache.async_get_cache(key=spend_key)
 
     async def async_filter_deployments(
         self,
diff --git a/litellm/proxy/management_helpers/access_group_key_sync.py b/litellm/proxy/management_helpers/access_group_key_sync.py
index 5d43cb29978..c9f93fae0d9 100644
--- a/litellm/proxy/management_helpers/access_group_key_sync.py
+++ b/litellm/proxy/management_helpers/access_group_key_sync.py
@@ -38,6 +38,7 @@ from litellm.proxy._types import (
 from litellm.proxy.auth.auth_checks import (
     _delete_cache_access_object,  # pyright: ignore[reportPrivateUsage]  # the access-group endpoints reach for this same cache primitive
 )
+from litellm.proxy.db.routing_prisma_wrapper import WriterPinnedClient
 from litellm.repositories.table_repositories import AccessGroupRepository
 
 
@@ -72,8 +73,9 @@ _REPOINT_KEY_SQL: Final = (
 
 
 def _raw_executor(prisma_client: object) -> _RawExecutor:
-    """Narrow the untyped Prisma client down to the raw-query call this module makes."""
-    return AccessGroupRepository(prisma_client).prisma_client.db  # pyright: ignore[reportAny]  # untyped Prisma client
+    """Narrow the untyped Prisma client down to the raw-query call this module makes, pinned to the writer."""
+    db: Final = AccessGroupRepository(prisma_client).prisma_client.db  # pyright: ignore[reportAny]  # untyped Prisma client
+    return WriterPinnedClient(db).db  # pyright: ignore[reportAny, reportReturnType]  # untyped Prisma client behind the pin
 
 
 async def _invalidate_access_group_cache(access_group_id: str) -> None:
diff --git a/litellm/proxy/proxy_server.py b/litellm/proxy/proxy_server.py
index 77a80ea0052..d52f05f6166 100644
--- a/litellm/proxy/proxy_server.py
+++ b/litellm/proxy/proxy_server.py
@@ -39,7 +39,7 @@ from typing import (
 import anyio
 import websockets
 import websockets.exceptions
-from pydantic import BaseModel, Json, JsonValue, ValidationError
+from pydantic import BaseModel, Json, JsonValue, TypeAdapter, ValidationError
 from typing_extensions import NotRequired, ReadOnly, assert_never
 
 from litellm._uuid import uuid
@@ -60,6 +60,7 @@ from litellm.constants import (
     LITELLM_SETTINGS_SAFE_DB_OVERRIDES,
     LITELLM_UI_ALLOW_HEADERS,
     LITELLM_UI_SESSION_DURATION,
+    RUNTIME_UPDATABLE_ROUTER_SETTINGS,
 )
 from litellm.litellm_core_utils.litellm_logging import (
     _init_custom_logger_compatible_class,
@@ -253,6 +254,7 @@ from litellm.constants import (
     PROXY_BUDGET_RESCHEDULER_MAX_TIME,
     PROXY_BUDGET_RESCHEDULER_MIN_TIME,
     PROXY_CONFIG_RELOAD_INTERVAL_SECONDS,
+    ROUTER_SETTINGS_MANAGED_OUTSIDE_CONFIG,
     USER_SPEND_ALERTS_JOB_ID,
     WEEKLY_SPEND_REPORT_JOB_ID,
 )
@@ -352,7 +354,10 @@ from litellm.proxy.common_utils.load_config_utils import (
     get_file_contents_from_s3,
 )
 from litellm.proxy.common_utils.model_deprecation import collect_model_deprecations
-from litellm.proxy.common_utils.model_listing_utils import TeamModelNameTranslator
+from litellm.proxy.common_utils.model_listing_utils import (
+    TeamModelNameTranslator,
+    configured_display_names,
+)
 from litellm.proxy.common_utils.openai_endpoint_utils import (
     remove_sensitive_info_from_deployment,
 )
@@ -2275,7 +2280,7 @@ user_api_key_cache: UserApiKeyCache = UserApiKeyCache(
 )
 spend_counter_cache: Final = DualCache(default_in_memory_ttl=UserAPIKeyCacheTTLEnum.in_memory_cache_ttl.value)
 cli_sso_session_cache: Final = DualCache(default_in_memory_ttl=CLI_SSO_SESSION_TTL_SECONDS)
-model_max_budget_limiter: Final = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=user_api_key_cache)
+model_max_budget_limiter: Final = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=spend_counter_cache)
 litellm.logging_callback_manager.add_litellm_callback(model_max_budget_limiter)
 redis_usage_cache: RedisCache | None = None  # redis cache used for tracking spend, tpm/rpm limits
 polling_via_cache_enabled: Literal["all"] | list[str] | bool = False
@@ -5710,13 +5715,9 @@ class ProxyConfig:
         router_settings: Final = config.get("router_settings", None)
 
         if router_settings and isinstance(router_settings, dict):
-            # model list and search_tools already set
-            exclude_args: Final = {
-                "model_list",
-                "search_tools",
-            }
-
-            available_args: Final = [x for x in litellm.Router.get_valid_args() if x not in exclude_args]
+            available_args: Final = [
+                x for x in litellm.Router.get_valid_args() if x not in ROUTER_SETTINGS_MANAGED_OUTSIDE_CONFIG
+            ]
 
             for k, v in router_settings.items():
                 if k in available_args:
@@ -10223,7 +10224,8 @@ async def model_list(
         # The internal routing key drives the metadata/fallback lookup, while the
         # public name is what the client sees as the model id.
         model_data = []
-        for response_id, lookup_id in TeamModelNameTranslator.listing_entries(all_models, llm_router, settings):
+        admin_entries: Final = TeamModelNameTranslator.listing_entries(all_models, llm_router, settings)
+        for response_id, lookup_id in admin_entries:
             model_info = create_model_info_response(
                 model_id=lookup_id,
                 provider="openai",
@@ -10236,7 +10238,10 @@ async def model_list(
 
         if wants_anthropic_format:
             admin_listing: Final = cast(Sequence[ModelInfoResponse], model_data)  # cast-ok: rows built above
-            return create_anthropic_model_list_response(admin_listing)
+            return create_anthropic_model_list_response(
+                admin_listing,
+                display_names=configured_display_names(admin_entries, llm_router),
+            )
 
         return dict(
             data=model_data,
@@ -10267,7 +10272,8 @@ async def model_list(
     # The internal routing key drives the metadata/fallback lookup, while the
     # public name is what the client sees as the model id.
     model_data = []
-    for response_id, lookup_id in TeamModelNameTranslator.listing_entries(all_models, llm_router, settings):
+    entries: Final = TeamModelNameTranslator.listing_entries(all_models, llm_router, settings)
+    for response_id, lookup_id in entries:
         model_info = create_model_info_response(
             model_id=lookup_id,
             provider="openai",
@@ -10280,7 +10286,10 @@ async def model_list(
 
     if wants_anthropic_format:
         listing: Final = cast(Sequence[ModelInfoResponse], model_data)  # cast-ok: rows built above
-        return create_anthropic_model_list_response(listing)
+        return create_anthropic_model_list_response(
+            listing,
+            display_names=configured_display_names(entries, llm_router),
+        )
 
     return dict(
         data=model_data,
@@ -16207,6 +16216,7 @@ async def invitation_delete(
 )
 async def update_config(
     config_info: ConfigYAML,
+    request: Request,
     user_api_key_dict: UserAPIKeyAuth = Depends(user_api_key_auth),
 ):
     """
@@ -16222,6 +16232,26 @@ async def update_config(
         if user_api_key_dict.user_role != LitellmUserRoles.PROXY_ADMIN:
             raise HTTPException(status_code=403, detail="Only proxy admins can update config")
 
+        request_body: Final[Mapping[str, JsonValue]] = TypeAdapter(Mapping[str, JsonValue]).validate_python(
+            await request.json()
+        )
+        raw_router_settings: Final = request_body.get("router_settings")
+        if isinstance(raw_router_settings, dict):
+            supported_router_settings: Final = RUNTIME_UPDATABLE_ROUTER_SETTINGS | (
+                frozenset(litellm.Router.get_valid_args()) - ROUTER_SETTINGS_MANAGED_OUTSIDE_CONFIG
+            )
+            unsupported_router_settings: Final = sorted(set(raw_router_settings) - supported_router_settings)
+            if unsupported_router_settings:
+                raise HTTPException(
+                    status_code=400,
+                    detail={
+                        "error": (
+                            f"Unsupported router settings: {', '.join(unsupported_router_settings)} "
+                            "are not valid router settings"
+                        )
+                    },
+                )
+
         if prisma_client is None:
             raise Exception("No DB Connected")
 
@@ -16323,11 +16353,19 @@ async def update_config(
             )
 
         # router_settings: merge existing + request, request wins.
-        if config_info.router_settings is not None:
+        if isinstance(raw_router_settings, dict):
             existing = await _read_section("router_settings")
             before_router_settings: Final = copy.deepcopy(existing)
-            updates = config_info.router_settings.dict(exclude_none=True)
-            new_router_settings: Final = {**existing, **updates}
+            typed_router_settings: Final = (
+                config_info.router_settings.dict(exclude_none=True) if config_info.router_settings is not None else {}
+            )
+            raw_router_settings_without_none: Final = {
+                key: value
+                for key, value in raw_router_settings.items()
+                if key not in typed_router_settings and value is not None
+            }
+            router_settings_updates: Final = {**typed_router_settings, **raw_router_settings_without_none}
+            new_router_settings: Final = {**existing, **router_settings_updates}
             await _upsert_section("router_settings", new_router_settings)
             asyncio.create_task(
                 create_config_audit_log(
diff --git a/litellm/proxy/public_endpoints/agent_create_fields.json b/litellm/proxy/public_endpoints/agent_create_fields.json
index 36484cc1065..cc2fc17d759 100644
--- a/litellm/proxy/public_endpoints/agent_create_fields.json
+++ b/litellm/proxy/public_endpoints/agent_create_fields.json
@@ -107,7 +107,9 @@
         "required": true,
         "field_type": "text",
         "default_value": null,
-        "include_in_litellm_params": false
+        "include_in_litellm_params": false,
+        "validation_pattern": "^arn:aws[a-zA-Z0-9-]*:bedrock-agentcore:[a-z0-9-]+:[0-9]{12}:runtime/.+$",
+        "validation_message": "Enter the complete Bedrock AgentCore runtime ARN, including the runtime ID after \"runtime/\" (e.g. arn:aws:bedrock-agentcore:us-west-2:123456789012:runtime/my-agent-runtime)."
       }
     ],
     "litellm_params_template": {
diff --git a/litellm/proxy/rag_endpoints/endpoints.py b/litellm/proxy/rag_endpoints/endpoints.py
index db574f859b3..0ab7d99e4e4 100644
--- a/litellm/proxy/rag_endpoints/endpoints.py
+++ b/litellm/proxy/rag_endpoints/endpoints.py
@@ -9,6 +9,7 @@ Provides:
 import base64
 import json
 from collections.abc import Mapping
+from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, Final
 
 import orjson
@@ -19,6 +20,9 @@ from starlette.datastructures import UploadFile
 import litellm
 from litellm._logging import verbose_proxy_logger
 from litellm.constants import DEFAULT_MAX_RECURSE_DEPTH
+from litellm.integrations.vector_store_integrations.vector_store_pre_call_hook import (
+    LiteLLM_ManagedVectorStore,
+)
 from litellm.litellm_core_utils.streaming_handler import CustomStreamWrapper
 from litellm.proxy._types import *
 from litellm.proxy.auth.auth_utils import is_request_body_safe
@@ -36,6 +40,10 @@ from litellm.proxy.rag_endpoints.upload_security import (
     RejectedUpload,
     validate_upload,
 )
+from litellm.proxy.vector_store_endpoints.endpoints import (
+    build_request_data_from_managed_vector_store,
+    reject_caller_embedding_selection_params,
+)
 from litellm.proxy.vector_store_endpoints.utils import (
     assert_user_can_access_vector_store_id,
 )
@@ -120,12 +128,21 @@ def _collect_vector_store_ids_from_payload(payload: object) -> set[str]:
 async def _authorize_nested_vector_store_ids(
     payload: object,
     user_api_key_dict: UserAPIKeyAuth,
-) -> None:
-    for vector_store_id in sorted(_collect_vector_store_ids_from_payload(payload)):
-        await assert_user_can_access_vector_store_id(
-            vector_store_id=vector_store_id,
-            user_api_key_dict=user_api_key_dict,
-        )
+) -> Mapping[str, LiteLLM_ManagedVectorStore]:
+    """Authorize every nested vector store id and return the managed stores it resolved."""
+    return MappingProxyType(
+        {
+            vector_store_id: store
+            for vector_store_id in sorted(_collect_vector_store_ids_from_payload(payload))
+            if (
+                store := await assert_user_can_access_vector_store_id(
+                    vector_store_id=vector_store_id,
+                    user_api_key_dict=user_api_key_dict,
+                )
+            )
+            is not None
+        }
+    )
 
 
 def _build_file_metadata_entry(
@@ -700,11 +717,27 @@ async def rag_query(
                 status_code=400,
                 detail={"error": "retrieval_config must contain 'vector_store_id'"},
             )
-        await _authorize_nested_vector_store_ids(
+        reject_caller_embedding_selection_params(payload=retrieval_config, source="retrieval_config")
+        resolved_stores: Final = await _authorize_nested_vector_store_ids(
             payload=retrieval_config,
             user_api_key_dict=user_api_key_dict,
         )
 
+        # Merge litellm-managed vector store params (provider, region, embedding
+        # model, credentials, ...) from the registry: the same source the direct
+        # /vector_stores/{id}/search endpoint uses. Store-managed keys win on
+        # conflict so callers cannot override the store's provider or credentials.
+        managed_store: Final = resolved_stores.get(retrieval_config["vector_store_id"])
+        store_data: Final = (
+            await build_request_data_from_managed_vector_store(managed_store)
+            if managed_store is not None
+            else MappingProxyType({})
+        )
+        merged_retrieval_config: Final = {
+            **retrieval_config,
+            **store_data,
+        }  # mutable-ok: litellm.aquery requires a plain dict payload
+
         # Add litellm data
         request_data: dict[str, object] = {}
         request_data = await add_litellm_data_to_request(
@@ -716,13 +749,18 @@ async def rag_query(
             proxy_config=proxy_config,
         )
 
-        verbose_proxy_logger.debug("RAG Query - model: %s, retrieval_config: %s", model, retrieval_config)
+        verbose_proxy_logger.debug(
+            "RAG Query - model: %s, vector_store_id: %s, custom_llm_provider: %s",
+            model,
+            retrieval_config["vector_store_id"],
+            merged_retrieval_config.get("custom_llm_provider"),
+        )
 
         # Call query
         response: Final = await litellm.aquery(
             model=model,
             messages=messages,
-            retrieval_config=retrieval_config,
+            retrieval_config=merged_retrieval_config,
             rerank=rerank,
             stream=stream,
             router=llm_router,
diff --git a/litellm/proxy/vector_store_endpoints/endpoints.py b/litellm/proxy/vector_store_endpoints/endpoints.py
index a59d7a277cc..7d64e648e08 100644
--- a/litellm/proxy/vector_store_endpoints/endpoints.py
+++ b/litellm/proxy/vector_store_endpoints/endpoints.py
@@ -1,3 +1,5 @@
+from collections.abc import Mapping
+from types import MappingProxyType
 from typing import (
     Annotated,
     Any,  # noqa: TID251  # jsonify_object in proxy/utils.py is annotated with a bare dict
@@ -27,11 +29,69 @@ from litellm.types.vector_stores import IndexCreateRequest, IndexListResponse
 from litellm.vector_stores.vector_store_registry import VectorStoreIndexRegistry
 
 router: Final = APIRouter()
+
+BLOCKED_QUERY_EMBEDDING_SELECTION_PARAMS: Final = frozenset(
+    {
+        "embedding_model",
+        "litellm_embedding_model",
+        "litellm_embedding_config",
+        "litellm_credential_name",
+    }
+)
+
+
+def reject_caller_embedding_selection_params(payload: Mapping[str, object], source: str) -> None:
+    blocked: Final = sorted(BLOCKED_QUERY_EMBEDDING_SELECTION_PARAMS & payload.keys())
+    if blocked:
+        raise HTTPException(
+            status_code=400,
+            detail={
+                "error": f"'{blocked[0]}' cannot be set in {source}. "
+                "Embedding configuration comes from the vector store's server-side registration."
+            },
+        )
+
+
 ########################################################
 # OpenAI Compatible Endpoints
 ########################################################
 
 
+async def build_request_data_from_managed_vector_store(
+    vector_store: LiteLLM_ManagedVectorStore,
+) -> Mapping[str, object]:
+    """
+    Build request params (provider, credential ref, litellm_params) from an
+    already-resolved managed vector store.
+
+    ``litellm_embedding_config`` is resolved here, at request-handling time,
+    instead of at row-creation time: the resolved api_key/api_base/api_version
+    lives only in the returned per-request mapping and is never persisted back
+    to the registry cache. Legacy rows that already carry a resolved
+    (cleartext) config skip the lookup and pass through unchanged.
+    """
+    top_level: Final = MappingProxyType(
+        {
+            key: vector_store.get(key)
+            for key in ("custom_llm_provider", "litellm_credential_name")
+            if key in vector_store
+        }
+    )
+    litellm_params: Final = vector_store.get("litellm_params") or MappingProxyType({})
+    embedding_model: Final = litellm_params.get("litellm_embedding_model")
+    if not embedding_model or litellm_params.get("litellm_embedding_config"):
+        return MappingProxyType({**top_level, **litellm_params})
+
+    from litellm.proxy.proxy_server import prisma_client
+
+    resolved_config: Final = await _resolve_embedding_config(
+        embedding_model=embedding_model, prisma_client=prisma_client
+    )
+    if not resolved_config:
+        return MappingProxyType({**top_level, **litellm_params})
+    return MappingProxyType({**top_level, **litellm_params, "litellm_embedding_config": resolved_config})
+
+
 async def _update_request_data_with_litellm_managed_vector_store_registry(
     data: dict,
     vector_store_id: str,
@@ -51,47 +111,14 @@ async def _update_request_data_with_litellm_managed_vector_store_registry(
     vector_store_to_run: Final[LiteLLM_ManagedVectorStore | None] = await get_litellm_managed_vector_store(
         vector_store_id=vector_store_id
     )
-    if vector_store_to_run is not None:
-        if user_api_key_dict is not None:
-            await assert_user_can_access_vector_store(
-                vector_store=vector_store_to_run,
-                user_api_key_dict=user_api_key_dict,
-            )
-
-        if "custom_llm_provider" in vector_store_to_run:
-            data["custom_llm_provider"] = vector_store_to_run.get("custom_llm_provider")
-
-        if "litellm_credential_name" in vector_store_to_run:
-            data["litellm_credential_name"] = vector_store_to_run.get("litellm_credential_name")
-
-        if "litellm_params" in vector_store_to_run:
-            litellm_params = vector_store_to_run.get("litellm_params", {}) or {}
-            # Resolve ``litellm_embedding_config`` here, at request-handling
-            # time, instead of at row-creation time. The resolved
-            # ``api_key`` / ``api_base`` / ``api_version`` lives only in
-            # this per-request ``data`` dict and is never persisted.
-            # Legacy rows that already carry a resolved (cleartext)
-            # ``litellm_embedding_config`` skip the lookup and pass through
-            # unchanged so the embed call keeps working.
-            embedding_model: Final = litellm_params.get("litellm_embedding_model")
-            if embedding_model and not litellm_params.get("litellm_embedding_config"):
-                from litellm.proxy.proxy_server import prisma_client
-
-                resolved_config: Final = await _resolve_embedding_config(
-                    embedding_model=embedding_model, prisma_client=prisma_client
-                )
-                if resolved_config:
-                    # Build a fresh dict via spread instead of mutating
-                    # ``litellm_params`` in place — the registry hands back
-                    # a reference to its cached object, so an in-place
-                    # update would persist the resolved cleartext into the
-                    # in-memory cache for the lifetime of the process.
-                    litellm_params = {
-                        **litellm_params,
-                        "litellm_embedding_config": resolved_config,
-                    }
-            data.update(litellm_params)
-    return data
+    if vector_store_to_run is None:
+        return data
+    if user_api_key_dict is not None:
+        await assert_user_can_access_vector_store(
+            vector_store=vector_store_to_run,
+            user_api_key_dict=user_api_key_dict,
+        )
+    return {**data, **(await build_request_data_from_managed_vector_store(vector_store_to_run))}
 
 
 @router.post(
@@ -130,6 +157,7 @@ async def vector_store_search(
     )
 
     data = await _read_request_body(request=request)
+    reject_caller_embedding_selection_params(payload=data, source="the search request body")
     data["vector_store_id"] = vector_store_id
 
     # Check for legacy vector store registry (non-managed vector stores)
diff --git a/litellm/proxy/vector_store_endpoints/management_endpoints.py b/litellm/proxy/vector_store_endpoints/management_endpoints.py
index 183a03cc13c..244798ba05e 100644
--- a/litellm/proxy/vector_store_endpoints/management_endpoints.py
+++ b/litellm/proxy/vector_store_endpoints/management_endpoints.py
@@ -470,7 +470,7 @@ async def create_vector_store_in_db(
     # exposed every env-stored embedding-model credential on the
     # ``/vector_store/{new,info,update,list}`` responses. Keep the user's
     # raw ``litellm_embedding_model`` reference; resolution now happens in
-    # ``_update_request_data_with_litellm_managed_vector_store_registry``
+    # ``build_request_data_from_managed_vector_store``
     # at request-handling time so the cleartext config exists only in
     # per-request memory and never reaches the database.
     if litellm_params:
@@ -864,7 +864,7 @@ async def update_vector_store(
         # embedding-config auto-resolve previously persisted cleartext
         # credentials into the row; resolution now happens at request-
         # handling time in
-        # ``_update_request_data_with_litellm_managed_vector_store_registry``
+        # ``build_request_data_from_managed_vector_store``
         # so this row only ever stores the user-supplied
         # ``litellm_embedding_model`` reference.
         if "litellm_params" in update_data:
diff --git a/litellm/rag/main.py b/litellm/rag/main.py
index 7bc1a6a52a3..94bfc305a6a 100644
--- a/litellm/rag/main.py
+++ b/litellm/rag/main.py
@@ -14,6 +14,7 @@ import contextvars
 from collections.abc import Coroutine, Iterator
 from contextlib import contextmanager
 from functools import partial
+from types import MappingProxyType
 from typing import TYPE_CHECKING, Any, Final
 
 import httpx
@@ -50,6 +51,21 @@ INGESTION_REGISTRY: Final[dict[str, type[BaseRAGIngestion]]] = {
     "vertex_ai": VertexAIRAGIngestion,
 }
 
+# Only these retrieval_config keys are forwarded to vector_stores.asearch as
+# provider-specific params. The explicit allowlist keeps caller-controlled
+# connection overrides (api_base, api_key, ...) away from the search call,
+# where they could redirect store credentials to an attacker-chosen host.
+_FORWARDABLE_RETRIEVAL_CONFIG_KEYS: Final = frozenset(
+    {
+        "aws_region_name",
+        "vector_bucket_name",
+        "embedding_model",
+        "litellm_embedding_model",
+        "litellm_embedding_config",
+        "litellm_credential_name",
+    }
+)
+
 
 def get_ingestion_class(provider: str) -> type[BaseRAGIngestion]:
     """
@@ -224,13 +240,20 @@ async def _execute_query_pipeline(
         raise ValueError("No query found in messages for RAG query")
 
     # 2. Search vector store
+    # Forward allowlisted provider retrieval_config extras (region, embedding
+    # model, bucket, credential refs) to the search call; kwargs win on conflict.
+    provider_search_params: Final = MappingProxyType(
+        {k: v for k, v in retrieval_config.items() if k in _FORWARDABLE_RETRIEVAL_CONFIG_KEYS}
+    )
+    forwarded_search_params: Final = MappingProxyType({**provider_search_params, **kwargs})
     with _suppressed_sub_call_billing():
         search_response: Final = await litellm.vector_stores.asearch(
             vector_store_id=retrieval_config["vector_store_id"],
             query=query_text,
             max_num_results=retrieval_config.get("top_k", 10),
             custom_llm_provider=retrieval_config.get("custom_llm_provider", "openai"),
-            **kwargs,
+            router=router,
+            **forwarded_search_params,
         )
 
     search_provider: Final = retrieval_config.get("custom_llm_provider", "openai")
diff --git a/litellm/rerank_api/main.py b/litellm/rerank_api/main.py
index c8f7842aebf..37ca989b8d3 100644
--- a/litellm/rerank_api/main.py
+++ b/litellm/rerank_api/main.py
@@ -6,6 +6,7 @@ from typing import Any, Final, Literal
 
 import litellm
 from litellm._logging import verbose_logger
+from litellm.litellm_core_utils.get_llm_provider_logic import declared_authenticating_provider
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
 from litellm.llms.base_llm.rerank.transformation import BaseRerankConfig
 from litellm.llms.bedrock.rerank.handler import BedrockRerankHandler
@@ -43,10 +44,23 @@ async def arerank(
     """
     Async: Reranks a list of documents based on their relevance to the query
     """
+    _custom_llm_provider: str | None = (
+        None  # rebind-ok: set by the declared-provider guard or the get_llm_provider unpack; read in the except
+    )
     try:
         loop: Final = asyncio.get_event_loop()
         kwargs["arerank"] = True
 
+        declared_provider: Final = declared_authenticating_provider(model, custom_llm_provider)
+        if declared_provider is not None:
+            _custom_llm_provider = declared_provider  # rebind-ok: see pre-declaration above
+        else:
+            _, _custom_llm_provider, _, _ = litellm.get_llm_provider(  # rebind-ok: see pre-declaration above
+                model=model,
+                custom_llm_provider=custom_llm_provider,
+                api_base=kwargs.get("api_base", None),
+            )
+
         func: Final = partial(
             rerank,
             model,
@@ -70,7 +84,11 @@ async def arerank(
             response = init_response
         return response
     except Exception as e:
-        raise e
+        raise exception_type(
+            model=model,
+            custom_llm_provider=_custom_llm_provider or custom_llm_provider,
+            original_exception=e,
+        )
 
 
 @client
@@ -115,6 +133,7 @@ def rerank(
     model_info: Final = kwargs.get("model_info", None)
     user: Final = kwargs.get("user", None)
     client: Final = kwargs.get("client", None)
+    _custom_llm_provider: str | None = None  # rebind-ok: set by the get_llm_provider unpack; read in the except
     try:
         _is_async: Final = kwargs.pop("arerank", False) is True
         optional_params: Final = GenericLiteLLMParams(**kwargs)
@@ -127,7 +146,7 @@ def rerank(
 
         (
             model,
-            _custom_llm_provider,
+            _custom_llm_provider,  # rebind-ok: see pre-declaration above
             dynamic_api_key,
             dynamic_api_base,
         ) = litellm.get_llm_provider(
@@ -538,4 +557,8 @@ def rerank(
         return response
     except Exception as e:
         verbose_logger.error("Error in rerank: %s", e)
-        raise exception_type(model=model, custom_llm_provider=custom_llm_provider, original_exception=e)
+        raise exception_type(
+            model=model,
+            custom_llm_provider=_custom_llm_provider or custom_llm_provider,
+            original_exception=e,
+        )
diff --git a/litellm/responses/litellm_completion_transformation/streaming_iterator.py b/litellm/responses/litellm_completion_transformation/streaming_iterator.py
index db1c3acbefb..bc25f4fffb1 100644
--- a/litellm/responses/litellm_completion_transformation/streaming_iterator.py
+++ b/litellm/responses/litellm_completion_transformation/streaming_iterator.py
@@ -1169,16 +1169,6 @@ class LiteLLMCompletionStreamingIterator(ResponsesAPIStreamingIterator):
 
     def _emit_response_completed_event(self, litellm_model_response: ModelResponse) -> ResponseCompletedEvent | None:
         if litellm_model_response:
-            # Add cost to usage object if include_cost_in_streaming_usage is True
-            if litellm.include_cost_in_streaming_usage and self.litellm_logging_obj is not None:
-                usage: Final[object] = getattr(litellm_model_response, "usage", None)
-                if usage is not None:
-                    setattr(
-                        usage,
-                        "cost",
-                        self.litellm_logging_obj._response_cost_calculator(result=litellm_model_response),
-                    )
-
             # Transform the response
             responses_api_response: Final = (
                 LiteLLMCompletionResponsesConfig.transform_chat_completion_response_to_responses_api_response(
diff --git a/litellm/responses/streaming_iterator.py b/litellm/responses/streaming_iterator.py
index 82abac3e772..7871c85220c 100644
--- a/litellm/responses/streaming_iterator.py
+++ b/litellm/responses/streaming_iterator.py
@@ -407,23 +407,7 @@ class BaseResponsesAPIStreamingIterator:
                     openai_types.ResponsesAPIStreamEvents.RESPONSE_FAILED,
                 ):
                     self.completed_response = openai_responses_api_chunk
-                    # Add cost to usage object if include_cost_in_streaming_usage is True
-                    if litellm.include_cost_in_streaming_usage and self.logging_obj is not None:
-                        response_obj: Final[ResponsesAPIResponse | None] = getattr(
-                            openai_responses_api_chunk, "response", None
-                        )
-                        if response_obj:
-                            usage_obj: Final[ResponseAPIUsage | None] = getattr(response_obj, "usage", None)
-                            if usage_obj is not None:
-                                try:
-                                    cost: Final[float | None] = self.logging_obj._response_cost_calculator(
-                                        result=response_obj
-                                    )
-                                    if cost is not None:
-                                        setattr(usage_obj, "cost", cost)
-                                except Exception:
-                                    # Best-effort usage cost annotation should not break stream replay.
-                                    pass
+                    _stamp_responses_usage_cost(getattr(openai_responses_api_chunk, "response", None), self.logging_obj)
 
                     if _chunk_type == openai_types.ResponsesAPIStreamEvents.RESPONSE_FAILED:
                         self._handle_logging_failed_response()
@@ -1274,6 +1258,24 @@ def _add_text_like_part_events(
         )
 
 
+def _stamp_responses_usage_cost(
+    response_obj: ResponsesAPIResponse | None, logging_obj: LiteLLMLoggingObj | None
+) -> None:
+    if response_obj is None or logging_obj is None:
+        return
+    usage_obj: Final[ResponseAPIUsage | None] = getattr(response_obj, "usage", None)
+    if usage_obj is None:
+        return
+    if isinstance(getattr(usage_obj, "cost", None), (int, float)):
+        return
+    try:
+        cost: Final[float | None] = logging_obj._response_cost_calculator(result=response_obj)
+    except Exception:
+        return
+    if isinstance(cost, (int, float)) and cost > 0:
+        setattr(usage_obj, "cost", cost)
+
+
 def build_synthetic_response_events(
     *,
     transformed: ResponsesAPIResponse,
@@ -1281,15 +1283,7 @@ def build_synthetic_response_events(
     chunk_size: int,
 ) -> list[ResponsesAPIStreamingResponse]:
     openai_types: Final = _get_openai_response_types()
-    if litellm.include_cost_in_streaming_usage and logging_obj is not None:
-        usage_obj: Final = transformed.usage if hasattr(transformed, "usage") else None
-        if usage_obj is not None:
-            try:
-                cost: Final[float | None] = logging_obj._response_cost_calculator(result=transformed)
-                if cost is not None:
-                    setattr(usage_obj, "cost", cost)
-            except Exception:
-                pass
+    _stamp_responses_usage_cost(transformed, logging_obj)
 
     events: Final[list[ResponsesAPIStreamingResponse]] = [
         _build_response_status_event(openai_types.ResponsesAPIStreamEvents.RESPONSE_CREATED, transformed),
diff --git a/litellm/router.py b/litellm/router.py
index 23d8907fb49..b1285f9e55e 100644
--- a/litellm/router.py
+++ b/litellm/router.py
@@ -50,6 +50,7 @@ from litellm.constants import (
     DEFAULT_HEALTH_CHECK_INTERVAL,
     DEFAULT_HEALTH_CHECK_STALENESS_MULTIPLIER,
     DEFAULT_MAX_LRU_CACHE_SIZE,
+    RUNTIME_UPDATABLE_ROUTER_SETTINGS,
     SESSION_DEPLOYMENT_AFFINITY_TTL_METADATA_KEY,
 )
 from litellm.integrations.custom_logger import CustomLogger
@@ -354,6 +355,13 @@ _PreRoutingStrategyT = TypeVar("_PreRoutingStrategyT")
 _ALIAS_PARAMS_NEVER_FORWARDED: Final = frozenset({"model", "api_base", "api_key", "api_version"})
 _ALIAS_MARKER_FORWARDED_PARAMS_KWARG: Final = "_alias_marker_forwarded_params"
 
+_RUNTIME_TOGGLEABLE_PRE_CALL_CHECKS: Final[Mapping[str, type[CustomLogger]]] = MappingProxyType(
+    {
+        "prompt_caching": PromptCachingDeploymentCheck,
+        "enforce_model_rate_limits": ModelRateLimitingCheck,
+    }
+)
+
 
 def _stream_chunks_have_generated_content(chunks: Sequence[ModelResponseStream]) -> bool:
     for chunk in chunks:
@@ -2072,11 +2080,39 @@ class Router:
             if _callback is None:
                 continue
 
+            if self.optional_callbacks is not None and any(
+                isinstance(callback, type(_callback)) for callback in self.optional_callbacks
+            ):
+                continue
             if self.optional_callbacks is None:
                 self.optional_callbacks = []
             self.optional_callbacks.append(_callback)
             litellm.logging_callback_manager.add_litellm_callback(_callback)
 
+    def set_optional_pre_call_checks(self, optional_pre_call_checks: OptionalPreCallChecks | None) -> None:
+        if optional_pre_call_checks is None:
+            return
+        requested: Final = frozenset(optional_pre_call_checks)
+        for name, callback_cls in _RUNTIME_TOGGLEABLE_PRE_CALL_CHECKS.items():
+            if name not in requested:
+                self._remove_optional_callbacks_of_type(callback_cls)
+        self.add_optional_pre_call_checks(optional_pre_call_checks)
+
+    def _remove_optional_callbacks_of_type(self, callback_cls: type[CustomLogger]) -> None:
+        if self.optional_callbacks is None or not any(type(cb) is callback_cls for cb in self.optional_callbacks):
+            return
+        self.optional_callbacks = [cb for cb in self.optional_callbacks if type(cb) is not callback_cls]
+        if any(
+            router is not self and any(type(cb) is callback_cls for cb in (router.optional_callbacks or []))
+            for router in tuple(_live_routers)
+        ):
+            return
+        for cb in tuple(litellm.callbacks):
+            if type(cb) is callback_cls:
+                litellm.logging_callback_manager.remove_callback_from_list_by_object(
+                    litellm.callbacks, cb, require_self=False
+                )
+
     def print_deployment(self, deployment: dict):
         """
         returns a copy of the deployment with the api key masked
@@ -2324,7 +2360,7 @@ class Router:
     @overload
     async def acompletion(
         self, model: str, messages: list[AllMessageValues], stream: Literal[True, False] = False, **kwargs
-    ) -> CustomStreamWrapper | ModelResponse: 
+    ) -> CustomStreamWrapper | ModelResponse:
         ...
 
     # fmt: on
@@ -6374,8 +6410,6 @@ class Router:
             "responses",
             "generate_content",
             "generate_content_stream",
-            "vector_store_search",
-            "vector_store_create",
             "ocr",
             "search",
             "video_generation",
@@ -6399,6 +6433,8 @@ class Router:
             return sync_wrapper
 
         if call_type in (
+            "vector_store_search",
+            "vector_store_create",
             "vector_store_retrieve",
             "vector_store_list",
             "vector_store_update",
@@ -6410,11 +6446,16 @@ class Router:
                 client: object | None = None,
                 **kwargs,
             ):
-                if custom_llm_provider and "custom_llm_provider" not in kwargs:
-                    kwargs["custom_llm_provider"] = custom_llm_provider
-                if kwargs.get("model"):
-                    return self._generic_api_call_with_fallbacks(original_function=original_function, **kwargs)
-                return original_function(**kwargs)
+                provider_kwargs: Final = (
+                    MappingProxyType({**kwargs, "custom_llm_provider": custom_llm_provider})
+                    if custom_llm_provider and "custom_llm_provider" not in kwargs
+                    else MappingProxyType(kwargs)
+                )
+                if provider_kwargs.get("model"):
+                    return self._generic_api_call_with_fallbacks(original_function=original_function, **provider_kwargs)
+                if call_type == "vector_store_search":
+                    return original_function(**MappingProxyType({**provider_kwargs, "router": self}))
+                return original_function(**provider_kwargs)
 
             return vector_store_sync_wrapper
 
@@ -6590,6 +6631,7 @@ class Router:
                 return await self._init_vector_store_api_endpoints(
                     original_function=original_function,
                     custom_llm_provider=custom_llm_provider,
+                    call_type=call_type,
                     **kwargs,
                 )
             elif call_type in ("afile_delete", "afile_content"):
@@ -6630,6 +6672,7 @@ class Router:
         self,
         original_function: Callable,
         custom_llm_provider: str | None = None,
+        call_type: str | None = None,
         **kwargs,
     ):
         """
@@ -6648,6 +6691,13 @@ class Router:
                 **kwargs,
             )
 
+        # For search, pass the router so provider transforms can resolve
+        # router-managed embedding models (e.g. S3 Vectors query embeddings).
+        # The merge also overrides any client-supplied `router` key.
+        if call_type == "avector_store_search":
+            search_kwargs: Final = MappingProxyType({**kwargs, "router": self})
+            return await original_function(**search_kwargs)
+
         # Otherwise, call the original function directly
         return await original_function(**kwargs)
 
@@ -6664,7 +6714,10 @@ class Router:
         metadata. When present, decode the ID, replace ``container_id`` with the
         upstream value, and route through ``_ageneric_api_call_with_fallbacks`` so
         deployment credentials (e.g. regional ``api_base`` for Azure) match
-        :meth:`_init_responses_api_endpoints`. Otherwise call the handler directly.
+        :meth:`_init_responses_api_endpoints`. Create/list calls carry no container ID, so
+        they route through the deployment named by ``model`` when the caller passes one,
+        falling back to the direct call when no deployment matches. Otherwise call the
+        handler directly with global provider credentials.
         """
         if custom_llm_provider and "custom_llm_provider" not in kwargs:
             kwargs["custom_llm_provider"] = custom_llm_provider
@@ -6696,6 +6749,14 @@ class Router:
                     **kwargs,
                 )
 
+        requested_model: Final = kwargs.get("model")
+        if isinstance(requested_model, str) and requested_model.strip():
+            return await self._ageneric_api_call_with_fallbacks(
+                original_function=original_function,
+                passthrough_on_no_deployment=True,
+                **kwargs,
+            )
+
         return await original_function(**kwargs)
 
     async def _init_responses_api_endpoints(
@@ -9746,6 +9807,26 @@ class Router:
             coerce_token_limit(model_info.get("max_output_tokens")),
         )
 
+    def get_configured_display_name(self, model_name: str) -> "str | None":
+        """
+        Return the display_name explicitly configured in a concrete deployment's
+        model_info for model_name, via O(1) index lookup.
+
+        Returns None for wildcard-expanded or unknown names, and treats a
+        non-string or empty configured value as absent rather than failing the
+        listing. Like get_configured_token_limits, this never triggers pattern
+        matching or deep copies, so it is safe to call per listed model on the
+        /v1/models hot path.
+        """
+        deployment: Final = self.get_deployment_by_model_group_name(model_group_name=model_name)
+        if deployment is None:
+            return None
+
+        display_name: Final = deployment.model_info.get("display_name")
+        if isinstance(display_name, str) and display_name.strip():
+            return display_name
+        return None
+
     def get_deployment_credentials_with_provider(
         self, model_id: str, team_id: str | None = None
     ) -> dict[str, Any] | None:
@@ -11331,27 +11412,6 @@ class Router:
         """
         Update the router settings.
         """
-        # only the following settings are allowed to be configured
-        _allowed_settings: Final = [
-            "routing_strategy_args",
-            "routing_strategy",
-            "routing_groups",
-            "allowed_fails",
-            "cooldown_time",
-            "num_retries",
-            "timeout",
-            "max_retries",
-            "retry_after",
-            "fallbacks",
-            "context_window_fallbacks",
-            "retry_policy",
-            "model_group_retry_policy",
-            "model_group_alias",
-            "enable_weighted_failover",
-            "enable_tag_filtering",
-            "tag_routing_prefix",
-        ]
-
         _int_settings: Final = [
             "timeout",
             "num_retries",
@@ -11364,13 +11424,15 @@ class Router:
         rebuild_routing_groups = False
         relink_lar1_from_args = False
         for var in kwargs:
-            if var in _allowed_settings:
+            if var in RUNTIME_UPDATABLE_ROUTER_SETTINGS:
                 if var in _int_settings:
                     _casted_value = int(kwargs[var])
                     setattr(self, var, _casted_value)
                 elif var == "routing_groups":
                     self._routing_groups_input = kwargs[var]
                     rebuild_routing_groups = True
+                elif var == "optional_pre_call_checks":
+                    self.set_optional_pre_call_checks(kwargs[var])
                 elif var == "retry_policy":
                     value = kwargs[var]
                     if isinstance(value, dict):
diff --git a/litellm/types/proxy/public_endpoints/public_endpoints.py b/litellm/types/proxy/public_endpoints/public_endpoints.py
index f6ee054ceaa..c7f80a61e0f 100644
--- a/litellm/types/proxy/public_endpoints/public_endpoints.py
+++ b/litellm/types/proxy/public_endpoints/public_endpoints.py
@@ -43,6 +43,8 @@ class AgentCredentialField(BaseModel):
     options: list[str] | None = None
     default_value: str | None = None
     include_in_litellm_params: bool | None = None
+    validation_pattern: str | None = None
+    validation_message: str | None = None
 
 
 class AgentCreateInfo(BaseModel):
diff --git a/litellm/types/router.py b/litellm/types/router.py
index e0957383aac..2a5f264cee3 100644
--- a/litellm/types/router.py
+++ b/litellm/types/router.py
@@ -106,6 +106,20 @@ class RetryPolicy(BaseModel):
     InternalServerErrorRetries: int | None = None
 
 
+OptionalPreCallChecks = list[
+    Literal[
+        "prompt_caching",
+        "router_budget_limiting",
+        "responses_api_deployment_check",
+        "deployment_affinity",
+        "session_affinity",
+        "forward_client_headers_by_model_group",
+        "enforce_model_rate_limits",
+        "encrypted_content_affinity",
+    ]
+]
+
+
 class UpdateRouterConfig(BaseModel):
     """
     Set of params that you can modify via `router.update_settings()`.
@@ -128,6 +142,7 @@ class UpdateRouterConfig(BaseModel):
     model_group_alias: dict[str, str | dict] | None = {}
     enable_tag_filtering: bool | None = None
     tag_routing_prefix: str | None = None
+    optional_pre_call_checks: OptionalPreCallChecks | None = None
 
     model_config = ConfigDict(protected_namespaces=())
 
@@ -869,20 +884,6 @@ class FallbackAccessCheck(Protocol):
     async def __call__(self, *, model: str, request_kwargs: Mapping[str, object], llm_router: "Router") -> bool: ...
 
 
-OptionalPreCallChecks = list[
-    Literal[
-        "prompt_caching",
-        "router_budget_limiting",
-        "responses_api_deployment_check",
-        "deployment_affinity",
-        "session_affinity",
-        "forward_client_headers_by_model_group",
-        "enforce_model_rate_limits",
-        "encrypted_content_affinity",
-    ]
-]
-
-
 class LiteLLM_RouterFileObject(TypedDict, total=False):
     """
     Tracking the litellm params hash, used for mapping the file id to the right model
diff --git a/litellm/vector_stores/main.py b/litellm/vector_stores/main.py
index 9b0ff71730a..cd576755f5f 100644
--- a/litellm/vector_stores/main.py
+++ b/litellm/vector_stores/main.py
@@ -7,7 +7,7 @@ import builtins
 import contextvars
 from collections.abc import Coroutine, Mapping
 from functools import partial
-from typing import Final
+from typing import TYPE_CHECKING, Final
 
 import httpx
 
@@ -29,6 +29,9 @@ from litellm.types.vector_stores import (
 from litellm.utils import ProviderConfigManager, client
 from litellm.vector_stores.utils import VectorStoreRequestUtils
 
+if TYPE_CHECKING:
+    from litellm.router import Router
+
 ####### ENVIRONMENT VARIABLES ###################
 # Initialize any necessary instances or variables here
 base_llm_http_handler = BaseLLMHTTPHandler()
@@ -280,6 +283,7 @@ async def asearch(
     timeout: float | httpx.Timeout | None = None,
     # LiteLLM specific params,
     custom_llm_provider: str | None = None,
+    router: "Router | None" = None,
     **kwargs,
 ) -> VectorStoreSearchResponse:
     """
@@ -308,6 +312,7 @@ async def asearch(
             extra_body=extra_body,
             timeout=timeout,
             custom_llm_provider=custom_llm_provider,
+            router=router,
             **kwargs,
         )
 
@@ -347,6 +352,7 @@ def search(
     timeout: float | httpx.Timeout | None = None,
     # LiteLLM specific params,
     custom_llm_provider: str | None = None,
+    router: "Router | None" = None,
     **kwargs,
 ) -> VectorStoreSearchResponse | Coroutine[object, object, VectorStoreSearchResponse]:
     """
@@ -450,6 +456,7 @@ def search(
             timeout=timeout or request_timeout,
             _is_async=_is_async,
             client=kwargs.get("client"),
+            router=router,
         )
 
         return response
diff --git a/model_prices_and_context_window.json b/model_prices_and_context_window.json
index a3cfb300ea6..2846d12db6e 100644
--- a/model_prices_and_context_window.json
+++ b/model_prices_and_context_window.json
@@ -9643,7 +9643,8 @@
         "supported_endpoints": [
             "/v1/images/generations",
             "/v1/images/edits"
-        ]
+        ],
+        "deprecation_date": "2026-10-01"
     },
     "azure_ai/MAI-Image-2.5-Flash": {
         "input_cost_per_image_token": 1.75e-06,
@@ -9656,7 +9657,8 @@
         "supported_endpoints": [
             "/v1/images/generations",
             "/v1/images/edits"
-        ]
+        ],
+        "deprecation_date": "2026-10-01"
     },
     "azure_ai/MAI-Image-2e": {
         "deprecation_date": "2026-08-15",
@@ -10155,7 +10157,9 @@
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_reasoning": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "cache_read_input_token_cost": 1.45e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/deepseek-v4-flash": {
         "deprecation_date": "2028-02-20",
@@ -10169,18 +10173,20 @@
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_reasoning": true,
-        "supports_tool_choice": true
-    },
-    "azure_ai/deepseek-v4-flash-0731": {
+        "supports_tool_choice": true,
         "cache_read_input_token_cost": 2.8e-08,
+        "supports_prompt_caching": true
+    },
+    "azure_ai/DeepSeek-V4-Flash-0731": {
+        "cache_read_input_token_cost": 1.4e-08,
         "deprecation_date": "2026-12-03",
-        "input_cost_per_token": 1.9e-07,
+        "input_cost_per_token": 4.4e-07,
         "litellm_provider": "azure_ai",
         "max_input_tokens": 1000000,
         "max_output_tokens": 128000,
         "max_tokens": 128000,
         "mode": "chat",
-        "output_cost_per_token": 5.1e-07,
+        "output_cost_per_token": 1.32e-06,
         "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/deepseek/",
         "supports_function_calling": true,
         "supports_prompt_caching": true,
@@ -10400,11 +10406,13 @@
         "max_tokens": 262144,
         "mode": "chat",
         "output_cost_per_token": 3e-06,
-        "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/kimi-k2-5-now-in-microsoft-foundry/4492321",
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
         "supports_function_calling": true,
         "supports_tool_choice": true,
         "supports_video_input": true,
-        "supports_vision": true
+        "supports_vision": true,
+        "cache_read_input_token_cost": 1e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/kimi-k2.6": {
         "deprecation_date": "2027-04-16",
@@ -10415,7 +10423,7 @@
         "max_tokens": 262144,
         "mode": "chat",
         "output_cost_per_token": 4e-06,
-        "source": "https://techcommunity.microsoft.com/blog/azure-ai-foundry-blog/introducing-kimi-k2-6-in-microsoft-foundry/4513125",
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
         "supported_modalities": [
             "text",
             "image"
@@ -10426,7 +10434,9 @@
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_tool_choice": true,
-        "supports_vision": true
+        "supports_vision": true,
+        "cache_read_input_token_cost": 1.6e-07,
+        "supports_prompt_caching": true
     },
     "azure_ai/ministral-3b": {
         "input_cost_per_token": 4e-08,
@@ -12110,7 +12120,7 @@
         "max_output_tokens": 2048,
         "max_tokens": 2048,
         "mode": "chat",
-        "output_cost_per_token": 2.65e-06,
+        "output_cost_per_token": 6e-07,
         "supports_pdf_input": true
     },
     "bedrock/us-west-1/meta.llama3-70b-instruct-v1:0": {
@@ -23514,6 +23524,63 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "vertex_ai/gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "vertex_ai",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "regional_endpoint_uplift_multiplier": 1.1,
+        "source": "https://cloud.google.com/vertex-ai/generative-ai/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "vertex_ai/gemini-3.1-pro-preview": {
         "prompt_cache_min_tokens": 4096,
         "cache_read_input_token_cost": 2e-07,
@@ -25351,6 +25418,65 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "gemini/gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "gemini",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "rpm": 2000,
+        "source": "https://ai.google.dev/gemini-api/docs/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_output": false,
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "tpm": 800000,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "gemini/gemini-omni-flash-preview": {
         "input_cost_per_audio_token": 1.5e-06,
         "input_cost_per_token": 1.5e-06,
@@ -25759,6 +25885,63 @@
         "web_search_billing_unit": "per_query",
         "google_maps_grounding_cost_per_query": 0.014
     },
+    "gemini-3.8-flash": {
+        "prompt_cache_min_tokens": 4096,
+        "cache_read_input_token_cost": 7.5e-08,
+        "cache_read_input_token_cost_flex": 3.75e-08,
+        "input_cost_per_token": 7.5e-07,
+        "input_cost_per_token_batches": 3.75e-07,
+        "input_cost_per_token_flex": 3.75e-07,
+        "litellm_provider": "vertex_ai-language-models",
+        "max_input_tokens": 1048576,
+        "max_output_tokens": 65536,
+        "max_tokens": 65536,
+        "mode": "chat",
+        "output_cost_per_reasoning_token": 3.75e-06,
+        "output_cost_per_token": 3.75e-06,
+        "output_cost_per_token_batches": 1.875e-06,
+        "output_cost_per_token_flex": 1.875e-06,
+        "source": "https://ai.google.dev/gemini-api/docs/pricing",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/completions",
+            "/v1/batch"
+        ],
+        "supported_modalities": [
+            "text",
+            "image",
+            "audio",
+            "video"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_audio_output": false,
+        "supports_audio_input": true,
+        "supports_function_calling": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_url_context": true,
+        "supports_video_input": true,
+        "supports_vision": true,
+        "supports_web_search": true,
+        "supports_native_streaming": true,
+        "input_cost_per_token_priority": 1.35e-06,
+        "output_cost_per_token_priority": 6.75e-06,
+        "cache_read_input_token_cost_priority": 1.35e-07,
+        "search_context_cost_per_query": {
+            "search_context_size_low": 0.014,
+            "search_context_size_medium": 0.014,
+            "search_context_size_high": 0.014
+        },
+        "web_search_billing_unit": "per_query",
+        "google_maps_grounding_cost_per_query": 0.014
+    },
     "gemini/gemini-2.5-pro-preview-tts": {
         "cache_read_input_token_cost": 1.25e-07,
         "input_cost_per_audio_token": 7e-07,
@@ -29098,16 +29281,19 @@
         "cache_creation_input_token_cost": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 1e-05,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
         "cache_creation_input_token_cost_flex": 2.5e-06,
         "cache_creation_input_token_cost_priority": 1e-05,
         "cache_read_input_token_cost": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens": 8e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 4e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
         "cache_read_input_token_cost_flex": 2e-07,
         "cache_read_input_token_cost_priority": 8e-07,
         "input_cost_per_token": 4e-06,
         "input_cost_per_token_above_272k_tokens": 8e-06,
         "input_cost_per_token_above_272k_tokens_flex": 4e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
         "input_cost_per_token_batches": 2e-06,
         "input_cost_per_token_flex": 2e-06,
         "input_cost_per_token_priority": 8e-06,
@@ -29119,6 +29305,7 @@
         "output_cost_per_token": 2e-05,
         "output_cost_per_token_above_272k_tokens": 3e-05,
         "output_cost_per_token_above_272k_tokens_flex": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
         "output_cost_per_token_batches": 1e-05,
         "output_cost_per_token_flex": 1e-05,
         "output_cost_per_token_priority": 4e-05,
@@ -29161,16 +29348,19 @@
         "cache_creation_input_token_cost": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 1e-05,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
         "cache_creation_input_token_cost_flex": 2.5e-06,
         "cache_creation_input_token_cost_priority": 1e-05,
         "cache_read_input_token_cost": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens": 8e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 4e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
         "cache_read_input_token_cost_flex": 2e-07,
         "cache_read_input_token_cost_priority": 8e-07,
         "input_cost_per_token": 4e-06,
         "input_cost_per_token_above_272k_tokens": 8e-06,
         "input_cost_per_token_above_272k_tokens_flex": 4e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
         "input_cost_per_token_batches": 2e-06,
         "input_cost_per_token_flex": 2e-06,
         "input_cost_per_token_priority": 8e-06,
@@ -29182,6 +29372,7 @@
         "output_cost_per_token": 2e-05,
         "output_cost_per_token_above_272k_tokens": 3e-05,
         "output_cost_per_token_above_272k_tokens_flex": 1.5e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
         "output_cost_per_token_batches": 1e-05,
         "output_cost_per_token_flex": 1e-05,
         "output_cost_per_token_priority": 4e-05,
@@ -29225,16 +29416,19 @@
         "cache_creation_input_token_cost": 2.5e-06,
         "cache_creation_input_token_cost_above_272k_tokens": 5e-06,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 2.5e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-05,
         "cache_creation_input_token_cost_flex": 1.25e-06,
         "cache_creation_input_token_cost_priority": 5e-06,
         "cache_read_input_token_cost": 2e-07,
         "cache_read_input_token_cost_above_272k_tokens": 4e-07,
         "cache_read_input_token_cost_above_272k_tokens_flex": 2e-07,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-07,
         "cache_read_input_token_cost_flex": 1e-07,
         "cache_read_input_token_cost_priority": 4e-07,
         "input_cost_per_token": 2e-06,
         "input_cost_per_token_above_272k_tokens": 4e-06,
         "input_cost_per_token_above_272k_tokens_flex": 2e-06,
+        "input_cost_per_token_above_272k_tokens_priority": 8e-06,
         "input_cost_per_token_batches": 1e-06,
         "input_cost_per_token_flex": 1e-06,
         "input_cost_per_token_priority": 4e-06,
@@ -29246,6 +29440,7 @@
         "output_cost_per_token": 1.2e-05,
         "output_cost_per_token_above_272k_tokens": 1.8e-05,
         "output_cost_per_token_above_272k_tokens_flex": 9e-06,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-05,
         "output_cost_per_token_batches": 6e-06,
         "output_cost_per_token_flex": 6e-06,
         "output_cost_per_token_priority": 2.4e-05,
@@ -29288,16 +29483,19 @@
         "cache_creation_input_token_cost": 2.5e-07,
         "cache_creation_input_token_cost_above_272k_tokens": 5e-07,
         "cache_creation_input_token_cost_above_272k_tokens_flex": 2.5e-07,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-06,
         "cache_creation_input_token_cost_flex": 1.25e-07,
         "cache_creation_input_token_cost_priority": 5e-07,
         "cache_read_input_token_cost": 2e-08,
         "cache_read_input_token_cost_above_272k_tokens": 4e-08,
         "cache_read_input_token_cost_above_272k_tokens_flex": 2e-08,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08,
         "cache_read_input_token_cost_flex": 1e-08,
         "cache_read_input_token_cost_priority": 4e-08,
         "input_cost_per_token": 2e-07,
         "input_cost_per_token_above_272k_tokens": 4e-07,
         "input_cost_per_token_above_272k_tokens_flex": 2e-07,
+        "input_cost_per_token_above_272k_tokens_priority": 8e-07,
         "input_cost_per_token_batches": 1e-07,
         "input_cost_per_token_flex": 1e-07,
         "input_cost_per_token_priority": 4e-07,
@@ -29309,6 +29507,7 @@
         "output_cost_per_token": 1.2e-06,
         "output_cost_per_token_above_272k_tokens": 1.8e-06,
         "output_cost_per_token_above_272k_tokens_flex": 9e-07,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-06,
         "output_cost_per_token_batches": 6e-07,
         "output_cost_per_token_flex": 6e-07,
         "output_cost_per_token_priority": 2.4e-06,
@@ -29548,7 +29747,10 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": false
+        "supports_minimal_reasoning_effort": false,
+        "input_cost_per_token_above_272k_tokens_flex": 5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07
     },
     "gpt-5.5-2026-04-23": {
         "cache_read_input_token_cost": 5e-07,
@@ -29602,7 +29804,10 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": true,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": false
+        "supports_minimal_reasoning_effort": false,
+        "input_cost_per_token_above_272k_tokens_flex": 5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07
     },
     "gpt-5.5-pro": {
         "cache_read_input_token_cost": 3e-06,
@@ -29751,7 +29956,10 @@
         "supports_none_reasoning_effort": true,
         "default_reasoning_effort": "none",
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07
     },
     "gpt-5.4-2026-03-05": {
         "cache_read_input_token_cost": 2.5e-07,
@@ -29800,7 +30008,10 @@
         "supports_none_reasoning_effort": true,
         "default_reasoning_effort": "none",
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07
     },
     "gpt-5.4-pro": {
         "cache_read_input_token_cost": 3e-06,
@@ -29849,7 +30060,9 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": false,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 3e-05,
+        "output_cost_per_token_above_272k_tokens_flex": 0.000135
     },
     "gpt-5.4-pro-2026-03-05": {
         "cache_read_input_token_cost": 3e-06,
@@ -29898,7 +30111,9 @@
         "supports_web_search": true,
         "supports_none_reasoning_effort": false,
         "supports_xhigh_reasoning_effort": true,
-        "supports_minimal_reasoning_effort": true
+        "supports_minimal_reasoning_effort": true,
+        "input_cost_per_token_above_272k_tokens_flex": 3e-05,
+        "output_cost_per_token_above_272k_tokens_flex": 0.000135
     },
     "gpt-5.4-mini": {
         "cache_read_input_token_cost": 7.5e-08,
@@ -30834,17 +31049,18 @@
     },
     "gpt-realtime-2": {
         "cache_creation_input_audio_token_cost": 4e-07,
+        "cache_read_input_audio_token_cost": 4e-07,
         "cache_read_input_token_cost": 4e-07,
         "input_cost_per_audio_token": 3.2e-05,
         "input_cost_per_image": 5e-06,
         "input_cost_per_token": 4e-06,
         "litellm_provider": "openai",
-        "max_input_tokens": 32000,
-        "max_output_tokens": 4096,
-        "max_tokens": 4096,
+        "max_input_tokens": 128000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
         "mode": "realtime",
         "output_cost_per_audio_token": 6.4e-05,
-        "output_cost_per_token": 1.6e-05,
+        "output_cost_per_token": 2.4e-05,
         "supported_endpoints": [
             "/v1/realtime"
         ],
@@ -30908,8 +31124,8 @@
         "input_cost_per_token": 6e-07,
         "litellm_provider": "openai",
         "max_input_tokens": 128000,
-        "max_output_tokens": 4096,
-        "max_tokens": 4096,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
         "mode": "realtime",
         "output_cost_per_audio_token": 2e-05,
         "output_cost_per_token": 2.4e-06,
@@ -30941,7 +31157,7 @@
         "input_cost_per_audio_token": 1e-05,
         "input_cost_per_token": 6e-07,
         "litellm_provider": "openai",
-        "max_input_tokens": 128000,
+        "max_input_tokens": 32000,
         "max_output_tokens": 4096,
         "max_tokens": 4096,
         "mode": "realtime",
@@ -33705,19 +33921,21 @@
         "source": "https://mistral.ai/pricing#api-pricing"
     },
     "mistral/magistral-medium-latest": {
-        "input_cost_per_token": 2e-06,
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.5e-06,
         "litellm_provider": "mistral",
-        "max_input_tokens": 40000,
-        "max_output_tokens": 40000,
-        "max_tokens": 40000,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 5e-06,
-        "source": "https://mistral.ai/news/magistral",
+        "output_cost_per_token": 7.5e-06,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04",
         "supports_assistant_prefill": true,
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/magistral-small-2506": {
         "deprecation_date": "2025-11-30",
@@ -33736,19 +33954,21 @@
         "supports_tool_choice": true
     },
     "mistral/magistral-small-latest": {
-        "input_cost_per_token": 5e-07,
+        "cache_read_input_token_cost": 1.5e-08,
+        "input_cost_per_token": 1.5e-07,
         "litellm_provider": "mistral",
-        "max_input_tokens": 40000,
-        "max_output_tokens": 40000,
-        "max_tokens": 40000,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 1.5e-06,
-        "source": "https://mistral.ai/pricing#api-pricing",
+        "output_cost_per_token": 6e-07,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-small-4-0-26-03",
         "supports_assistant_prefill": true,
         "supports_function_calling": true,
         "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/magistral-small-1-2-2509": {
         "deprecation_date": "2026-07-31",
@@ -33880,16 +34100,21 @@
         "supports_vision": true
     },
     "mistral/mistral-medium": {
-        "input_cost_per_token": 2.7e-06,
+        "cache_read_input_token_cost": 1.5e-07,
+        "input_cost_per_token": 1.5e-06,
         "litellm_provider": "mistral",
-        "max_input_tokens": 32000,
-        "max_output_tokens": 8191,
-        "max_tokens": 8191,
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
         "mode": "chat",
-        "output_cost_per_token": 8.1e-06,
+        "output_cost_per_token": 7.5e-06,
+        "source": "https://docs.mistral.ai/models/model-cards/mistral-medium-3-5-26-04",
         "supports_assistant_prefill": true,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
         "supports_response_schema": true,
-        "supports_tool_choice": true
+        "supports_tool_choice": true,
+        "supports_vision": true
     },
     "mistral/mistral-medium-2312": {
         "deprecation_date": "2025-06-16",
@@ -41340,13 +41565,13 @@
         "source": "https://docs.together.ai/docs/serverless-models"
     },
     "together_ai/Qwen/Qwen3.8-2.4T-A95B": {
-        "cache_read_input_token_cost": 5e-07,
-        "input_cost_per_token": 2.5e-06,
+        "cache_read_input_token_cost": 2.5e-07,
+        "input_cost_per_token": 2e-06,
         "litellm_provider": "together_ai",
         "max_input_tokens": 1010000,
         "max_tokens": 1010000,
         "mode": "chat",
-        "output_cost_per_token": 6.25e-06,
+        "output_cost_per_token": 6e-06,
         "source": "https://docs.together.ai/docs/serverless-models",
         "supports_prompt_caching": true
     },
@@ -41956,6 +42181,70 @@
         "supports_parallel_tool_use_config": true,
         "prompt_cache_min_tokens": 1024
     },
+    "us-gov.anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock_converse",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "us-gov.anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock_converse",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
     "au.anthropic.claude-haiku-4-5-20251001-v1:0": {
         "cache_creation_input_token_cost": 1.375e-06,
         "cache_creation_input_token_cost_above_1hr": 2.2e-06,
@@ -45849,6 +46138,26 @@
         "mode": "rerank",
         "output_cost_per_token": 0.0
     },
+    "voyage/rerank-3": {
+        "input_cost_per_token": 5e-08,
+        "litellm_provider": "voyage",
+        "max_input_tokens": 32000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
+        "mode": "rerank",
+        "output_cost_per_token": 0.0,
+        "source": "https://docs.voyageai.com/docs/pricing"
+    },
+    "voyage/rerank-3-lite": {
+        "input_cost_per_token": 2e-08,
+        "litellm_provider": "voyage",
+        "max_input_tokens": 32000,
+        "max_output_tokens": 32000,
+        "max_tokens": 32000,
+        "mode": "rerank",
+        "output_cost_per_token": 0.0,
+        "source": "https://docs.voyageai.com/docs/pricing"
+    },
     "voyage/voyage-2": {
         "input_cost_per_token": 1e-07,
         "litellm_provider": "voyage",
@@ -47100,6 +47409,27 @@
         "supports_vision": true,
         "supports_web_search": true
     },
+    "xai/grok-build-latest": {
+        "cache_read_input_token_cost": 3e-07,
+        "cache_read_input_token_cost_above_200k_tokens": 6e-07,
+        "input_cost_per_token": 2e-06,
+        "input_cost_per_token_above_200k_tokens": 4e-06,
+        "litellm_provider": "xai",
+        "max_input_tokens": 500000,
+        "max_output_tokens": 500000,
+        "max_tokens": 500000,
+        "mode": "chat",
+        "output_cost_per_token": 6e-06,
+        "output_cost_per_token_above_200k_tokens": 1.2e-05,
+        "source": "https://docs.x.ai/developers/models",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_web_search": true
+    },
     "xai/grok-4.6": {
         "cache_read_input_token_cost": 5e-07,
         "cache_read_input_token_cost_above_200k_tokens": 1e-06,
@@ -57420,6 +57750,34 @@
         "supports_tool_choice": true,
         "supports_vision": false
     },
+    "fireworks_ai/accounts/fireworks/models/glm-5p3-flash": {
+        "cache_read_input_token_cost": 3e-08,
+        "input_cost_per_token": 1.5e-07,
+        "litellm_provider": "fireworks_ai",
+        "max_input_tokens": 1048576,
+        "max_tokens": 1048576,
+        "mode": "chat",
+        "output_cost_per_token": 5e-07,
+        "source": "https://docs.fireworks.ai/serverless/pricing",
+        "supports_function_calling": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "fireworks_ai/accounts/fireworks/models/inkling": {
+        "cache_read_input_token_cost": 1.7e-07,
+        "input_cost_per_token": 1e-06,
+        "litellm_provider": "fireworks_ai",
+        "max_input_tokens": 1048576,
+        "max_tokens": 1048576,
+        "mode": "chat",
+        "output_cost_per_token": 4.05e-06,
+        "source": "https://fireworks.ai/models/fireworks/inkling",
+        "supports_function_calling": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
     "fireworks_ai/accounts/fireworks/models/qwen3-embedding-8b": {
         "input_cost_per_token": 1e-07,
         "output_cost_per_token": 0.0,
@@ -57479,5 +57837,542 @@
         "supported_endpoints": [
             "/v1/audio/transcriptions"
         ]
+    },
+    "scaleway/glm-5.2": {
+        "input_cost_per_token": 1.8e-06,
+        "litellm_provider": "scaleway",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 16384,
+        "max_tokens": 16384,
+        "mode": "chat",
+        "output_cost_per_token": 5.5e-06,
+        "source": "https://www.scaleway.com/en/pricing/model-as-a-service/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_vision": false
+    },
+    "scaleway/deepseek-v4-flash-0731": {
+        "cache_read_input_token_cost": 8e-08,
+        "input_cost_per_token": 4e-07,
+        "litellm_provider": "scaleway",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 8e-07,
+        "source": "https://www.scaleway.com/en/pricing/model-as-a-service/",
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_vision": false
+    },
+    "azure_ai/kimi-k2.7-code": {
+        "cache_read_input_token_cost": 1.9e-07,
+        "deprecation_date": "2026-10-03",
+        "input_cost_per_token": 9.5e-07,
+        "litellm_provider": "azure_ai",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 262144,
+        "max_tokens": 262144,
+        "mode": "chat",
+        "output_cost_per_token": 4e-06,
+        "source": "https://azure.microsoft.com/en-us/pricing/details/ai-foundry-models/kimi/",
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-nano-3-30b": {
+        "input_cost_per_token": 7.2e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 2.88e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_native_structured_output": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-nano-12b-v2": {
+        "input_cost_per_token": 2.4e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_system_messages": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-west-1/nvidia.nemotron-super-3-120b": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 7.8e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/openai.gpt-oss-20b-1:0": {
+        "input_cost_per_token": 8.4e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.6e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/openai.gpt-oss-120b-1:0": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-west-1/anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-nano-3-30b": {
+        "input_cost_per_token": 7.2e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 262144,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 2.88e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_native_structured_output": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-nano-12b-v2": {
+        "input_cost_per_token": 2.4e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 8192,
+        "max_tokens": 8192,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_system_messages": true,
+        "supports_vision": true
+    },
+    "bedrock/us-gov-east-1/nvidia.nemotron-super-3-120b": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 256000,
+        "max_output_tokens": 32768,
+        "max_tokens": 32768,
+        "mode": "chat",
+        "output_cost_per_token": 7.8e-07,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/openai.gpt-oss-20b-1:0": {
+        "input_cost_per_token": 8.4e-08,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3.6e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/openai.gpt-oss-120b-1:0": {
+        "input_cost_per_token": 1.8e-07,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 128000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 7.2e-07,
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-sonnet-5": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+        "input_cost_per_token": 2.4e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.2e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": false,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock/us-gov-east-1/anthropic.claude-opus-4-8": {
+        "bedrock_converse_supports_strict_tools": false,
+        "bedrock_output_config_effort_ceiling": "xhigh",
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+        "input_cost_per_token": 6e-06,
+        "litellm_provider": "bedrock",
+        "max_input_tokens": 1000000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 3e-05,
+        "prompt_cache_min_tokens": 1024,
+        "supports_adaptive_thinking": true,
+        "supports_assistant_prefill": false,
+        "supports_computer_use": true,
+        "supports_function_calling": true,
+        "supports_max_reasoning_effort": true,
+        "supports_mid_conversation_system": true,
+        "supports_native_structured_output": true,
+        "supports_output_config": true,
+        "supports_parallel_tool_use_config": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_sampling_params": false,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "supports_xhigh_reasoning_effort": true
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.6-terra": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 2.64e-06,
+        "input_cost_per_token_above_272k_tokens": 5.28e-06,
+        "cache_creation_input_token_cost": 3.3e-06,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-06,
+        "cache_read_input_token_cost": 2.64e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-07,
+        "output_cost_per_token": 1.584e-05,
+        "output_cost_per_token_above_272k_tokens": 2.376e-05
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.6-luna": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 2.64e-07,
+        "input_cost_per_token_above_272k_tokens": 5.28e-07,
+        "cache_creation_input_token_cost": 3.3e-07,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-07,
+        "cache_read_input_token_cost": 2.64e-08,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-08,
+        "output_cost_per_token": 1.584e-06,
+        "output_cost_per_token_above_272k_tokens": 2.376e-06
+    },
+    "bedrock_mantle/us-gov-west-1/openai.gpt-5.4": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 3.3e-06,
+        "cache_read_input_token_cost": 3.3e-07,
+        "output_cost_per_token": 1.98e-05
+    },
+    "bedrock_mantle/us-gov-west-1/xai.grok-4.3": {
+        "use_openai_responses_path": true,
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 131072,
+        "max_output_tokens": 16384,
+        "max_tokens": 16384,
+        "mode": "chat",
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/responses"
+        ],
+        "supports_function_calling": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "source": "https://aws.amazon.com/bedrock/pricing/",
+        "input_cost_per_token": 1.5e-06,
+        "output_cost_per_token": 3e-06,
+        "cache_read_input_token_cost": 2.4e-07
+    },
+    "bedrock_mantle/us-gov-east-1/openai.gpt-5.4": {
+        "litellm_provider": "bedrock_mantle",
+        "max_input_tokens": 1050000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "responses",
+        "use_openai_responses_path": true,
+        "supported_endpoints": [
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text"
+        ],
+        "supports_function_calling": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": true,
+        "input_cost_per_token": 3.3e-06,
+        "cache_read_input_token_cost": 3.3e-07,
+        "output_cost_per_token": 1.98e-05
+    },
+    "azure/us-gov/gpt-5.1": {
+        "cache_read_input_token_cost": 1.71875e-07,
+        "default_reasoning_effort": "none",
+        "input_cost_per_token": 1.71875e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 272000,
+        "max_output_tokens": 128000,
+        "max_tokens": 128000,
+        "mode": "chat",
+        "output_cost_per_token": 1.375e-05,
+        "supported_endpoints": [
+            "/v1/chat/completions",
+            "/v1/batch",
+            "/v1/responses"
+        ],
+        "supported_modalities": [
+            "text",
+            "image"
+        ],
+        "supported_output_modalities": [
+            "text",
+            "image"
+        ],
+        "supports_function_calling": true,
+        "supports_native_streaming": true,
+        "supports_none_reasoning_effort": true,
+        "supports_parallel_function_calling": true,
+        "supports_pdf_input": true,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_system_messages": true,
+        "supports_tool_choice": true,
+        "supports_vision": true
+    },
+    "azure/us-gov/o3-mini": {
+        "cache_read_input_token_cost": 7.57e-07,
+        "input_cost_per_token": 1.513e-06,
+        "litellm_provider": "azure",
+        "max_input_tokens": 200000,
+        "max_output_tokens": 100000,
+        "max_tokens": 100000,
+        "mode": "chat",
+        "output_cost_per_token": 6.05e-06,
+        "supports_prompt_caching": true,
+        "supports_reasoning": true,
+        "supports_response_schema": true,
+        "supports_tool_choice": true,
+        "supports_vision": false
+    },
+    "azure/us-gov/text-embedding-3-large": {
+        "input_cost_per_token": 1.63e-07,
+        "litellm_provider": "azure",
+        "max_input_tokens": 8191,
+        "max_tokens": 8191,
+        "mode": "embedding",
+        "output_cost_per_token": 0.0
+    },
+    "azure/us-gov/text-embedding-3-small": {
+        "input_cost_per_token": 2.5e-08,
+        "litellm_provider": "azure",
+        "max_input_tokens": 8191,
+        "max_tokens": 8191,
+        "mode": "embedding",
+        "output_cost_per_token": 0.0
+    },
+    "cloudflare/@cf/openai/whisper": {
+        "input_cost_per_second": 7.5e-06,
+        "litellm_provider": "cloudflare",
+        "mode": "audio_transcription",
+        "output_cost_per_second": 0.0,
+        "source": "https://developers.cloudflare.com/workers-ai/models/whisper/",
+        "supported_endpoints": [
+            "/v1/audio/transcriptions"
+        ]
+    },
+    "cloudflare/@cf/openai/whisper-large-v3-turbo": {
+        "input_cost_per_second": 8.5e-06,
+        "litellm_provider": "cloudflare",
+        "mode": "audio_transcription",
+        "output_cost_per_second": 0.0,
+        "source": "https://developers.cloudflare.com/workers-ai/models/whisper-large-v3-turbo/",
+        "supported_endpoints": [
+            "/v1/audio/transcriptions"
+        ]
     }
 }
diff --git a/pyproject.toml b/pyproject.toml
index 2866e27e84c..60162544612 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -161,7 +161,7 @@ proxy-runtime = [
     "mangum>=0.17.0,<1.0",
     "azure-ai-contentsafety>=1.0.0,<2.0",
     "azure-storage-file-datalake>=12.20.0,<13.0",
-    "pypdf>=6.12.0,<7.0",
+    "pypdf>=6.16.1,<7.0",
     "llm-sandbox>=0.3.39,<1.0",
     "detect-secrets>=1.5.0,<2.0",
 ]
@@ -292,7 +292,7 @@ exclude = [
 
 [tool.uv]
 constraint-dependencies = [
-    "tornado>=6.5.6",
+    "tornado>=6.5.8",
     "aiohttp>=3.14.2,<4.0",
     "packaging>=24.0",
     "soupsieve>=2.8.4",
diff --git a/rust-toolchain.toml b/rust-toolchain.toml
new file mode 100644
index 00000000000..a1598ccbb34
--- /dev/null
+++ b/rust-toolchain.toml
@@ -0,0 +1,4 @@
+[toolchain]
+channel = "1.98.0"
+profile = "minimal"
+components = ["rustfmt", "clippy"]
diff --git a/tests/e2e/test_junit_properties.py b/tests/e2e/test_junit_properties.py
index c0596177cc1..02c1413c840 100644
--- a/tests/e2e/test_junit_properties.py
+++ b/tests/e2e/test_junit_properties.py
@@ -24,25 +24,10 @@ from junit_properties import (
 )
 
 
-class FakeMarker:
-    def __init__(self, name: str, *args: object) -> None:
-        self.name = name
-        self.args = args
-
-
-class FakeItem:
-    """The three attributes junit_properties reads off a pytest Item."""
-
-    def __init__(
-        self, nodeid: str, location: tuple[str, int | None, str], markers: tuple[FakeMarker, ...] = ()
-    ) -> None:
-        self.nodeid = nodeid
-        self.location = location
-        self.user_properties: list[tuple[str, str]] = []
-        self._markers = markers
-
-    def iter_markers(self, name: str):
-        return (marker for marker in self._markers if marker.name == name)
+def collected_item(request: pytest.FixtureRequest, name: str) -> pytest.Item:
+    """The Item pytest collected for test ``name`` in this file: the real nodeid,
+    location and marker machinery the collection hook reads, as pytest built it."""
+    return next(item for item in request.session.items if item.path == request.path and item.name == name)
 
 
 def repo_root() -> Path | None:
@@ -109,22 +94,22 @@ class TestSourceFromLocation:
 
 
 class TestResultProperties:
-    def test_every_test_carries_package_covers_and_source(self) -> None:
-        item = FakeItem(
-            "logging/test_x.py::TestFoo::test_bar",
-            ("logging/test_x.py", 40, "TestFoo.test_bar"),
-            (FakeMarker("covers", "LOG-1", "LOG-2"),),
-        )
-        assert result_properties(item) == (
-            ("package", "logging"),
+    def test_every_test_carries_package_covers_and_source(self, request: pytest.FixtureRequest) -> None:
+        """Read off this test's own collected Item, so the nodeid and location are
+        whatever pytest reports for the launch shape in use, and the marker is added
+        at run time so the coverage registry's collect-only pass never sees it."""
+        test = type(self).test_every_test_carries_package_covers_and_source
+        request.applymarker(pytest.mark.covers("LOG-1", "LOG-2"))
+        assert result_properties(collected_item(request, test.__name__)) == (
+            ("package", "root"),
             ("covers", "LOG-1,LOG-2"),
-            ("source", "tests/e2e/logging/test_x.py:41"),
+            ("source", f"tests/e2e/test_junit_properties.py:{test.__code__.co_firstlineno}"),
         )
 
-    def test_attach_is_idempotent(self) -> None:
+    def test_attach_is_idempotent(self, request: pytest.FixtureRequest) -> None:
         """Collection can run the hook more than once; a second pass must not
         double the <property> entries in the report."""
-        item = FakeItem("logging/test_x.py::test_bar", ("logging/test_x.py", 40, "test_bar"))
+        item = collected_item(request, type(self).test_attach_is_idempotent.__name__)
         attach_result_properties(item)
         attach_result_properties(item)
         assert [name for name, _ in item.user_properties] == ["package", "covers", "source"]
diff --git a/tests/ocr_tests/test_ocr_vertex_ai.py b/tests/ocr_tests/test_ocr_vertex_ai.py
index 1ba5b9d0883..1842eb063a5 100644
--- a/tests/ocr_tests/test_ocr_vertex_ai.py
+++ b/tests/ocr_tests/test_ocr_vertex_ai.py
@@ -5,9 +5,11 @@ Note: Vertex AI OCR automatically converts URLs to base64 data URIs since
 the Vertex AI endpoint doesn't have internet access.
 """
 
-import os
 import json
+import os
 import tempfile
+from typing import Final
+
 import pytest
 from base_ocr_unit_tests import BaseOCRTest
 
@@ -139,3 +141,19 @@ def test_vertex_ai_ocr_routing():
     assert isinstance(
         deepseek_variant, VertexAIDeepSeekOCRConfig
     ), "DeepSeek variant should route to VertexAIDeepSeekOCRConfig"
+
+
+@pytest.mark.parametrize("model", ("deepseek-ocr-maas", "deepseek-ai/deepseek-ocr-maas"))
+def test_deepseek_request_uses_single_provider_namespace(model: str) -> None:
+    from litellm.llms.vertex_ai.ocr.deepseek_transformation import (
+        VertexAIDeepSeekOCRConfig,
+    )
+
+    request: Final = VertexAIDeepSeekOCRConfig().transform_ocr_request(
+        model=model,
+        document={"type": "image_url", "image_url": "data:image/png;base64,AA=="},
+        optional_params={},
+        headers={},
+    )
+
+    assert request.data["model"] == "deepseek-ai/deepseek-ocr-maas"
diff --git a/tests/proxy_migration_tests/test_prisma_toolchain.py b/tests/proxy_migration_tests/test_prisma_toolchain.py
index d12a2c4dd4e..733870f3239 100644
--- a/tests/proxy_migration_tests/test_prisma_toolchain.py
+++ b/tests/proxy_migration_tests/test_prisma_toolchain.py
@@ -7,26 +7,36 @@ attempt fails identically. These tests pin the two behaviours that keep a
 container recoverable: an incomplete cache is deleted before Prisma is
 invoked, and the install gets a budget of its own rather than sharing the one
 that bounds each migration command.
+
+``prisma migrate deploy`` gets a budget of its own for the same reason: its
+runtime grows with the number of pending migrations, so a fresh database that
+replays every migration overran the per-command budget on slow machines and
+the proxy gave up after four identical timeouts.
 """
 
 import ast
 import json
+import logging
 import os
 import sys
 import time
+from collections.abc import Callable
 from pathlib import Path
 
 import pytest
 
 from litellm_proxy_extras.prisma_toolchain import (
     DEFAULT_PRISMA_COMMAND_TIMEOUT,
+    DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT,
     PRISMA_BOOTSTRAP_TIMEOUT_ENV_VAR,
     PRISMA_COMMAND_TIMEOUT_ENV_VAR,
+    PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR,
     ensure_prisma_toolchain,
     heal_incomplete_nodeenv_cache,
     node_binary_path,
     prisma_bootstrap_timeout,
     prisma_command_timeout,
+    prisma_migrate_deploy_timeout,
 )
 from litellm_proxy_extras.utils import ProxyExtrasDBManager
 
@@ -42,14 +52,27 @@ import time
 
 args = sys.argv[1:]
 cache_dir = os.environ["PRISMA_NODEENV_CACHE_DIR"]
-with pathlib.Path(os.environ["FAKE_PRISMA_LOG"]).open("a") as log:
+log_path = pathlib.Path(os.environ["FAKE_PRISMA_LOG"])
+earlier_same_command = sum(
+    1
+    for line in (log_path.read_text().splitlines() if log_path.exists() else [])
+    if json.loads(line)["args"][:2] == args[:2]
+)
+with log_path.open("a") as log:
     log.write(
         json.dumps({{"args": args, "cache_dir_present": os.path.isdir(cache_dir)}})
         + "\\n"
     )
 time.sleep(float(os.environ.get("FAKE_PRISMA_SLEEP", "0")))
 if args[:2] == ["migrate", "deploy"]:
+    if earlier_same_command == 0:
+        time.sleep(float(os.environ.get("FAKE_PRISMA_FIRST_DEPLOY_SLEEP", "0")))
+    elif os.environ.get("FAKE_PRISMA_LATER_DEPLOY_STDERR"):
+        print(os.environ["FAKE_PRISMA_LATER_DEPLOY_STDERR"], file=sys.stderr)
+        sys.exit(1)
     print("No pending migrations to apply")
+if args[:2] == ["db", "push"] and earlier_same_command == 0:
+    time.sleep(float(os.environ.get("FAKE_PRISMA_FIRST_PUSH_SLEEP", "0")))
 sys.exit(0)
 """
 
@@ -80,9 +103,14 @@ def toolchain_env(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> tuple[Path
     monkeypatch.setenv("PATH", f"{bin_dir}{os.pathsep}{os.environ['PATH']}")
     monkeypatch.delenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, raising=False)
     monkeypatch.delenv(PRISMA_BOOTSTRAP_TIMEOUT_ENV_VAR, raising=False)
+    monkeypatch.delenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, raising=False)
     return cache_dir, log_path
 
 
+def _deploy_calls(log_path: Path) -> list[list[str]]:
+    return [call["args"] for call in _fake_prisma_calls(log_path) if call["args"][:2] == ["migrate", "deploy"]]
+
+
 def _make_incomplete_cache(cache_dir: Path) -> None:
     (cache_dir / "lib").mkdir(parents=True)
     (cache_dir / "bin").mkdir()
@@ -209,25 +237,111 @@ def test_setup_database_prepares_the_toolchain_before_migrating(
     assert calls[0]["cache_dir_present"] is False
 
 
+@pytest.mark.parametrize("use_v2_resolver", [False, True], ids=["v1", "v2"])
+def test_migrate_deploy_is_not_bounded_by_the_per_command_timeout(
+    toolchain_env: tuple[Path, Path],
+    monkeypatch: pytest.MonkeyPatch,
+    use_v2_resolver: bool,
+) -> None:
+    """A fresh database replays every migration, which takes longer than any bookkeeping command."""
+    _, log_path = toolchain_env
+    monkeypatch.setenv("DATABASE_URL", "postgresql://u:p@localhost:9/x")
+    monkeypatch.setenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, "1")
+    monkeypatch.setenv("FAKE_PRISMA_FIRST_DEPLOY_SLEEP", "3")
+
+    assert ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=use_v2_resolver) is True
+    assert _deploy_calls(log_path) == [["migrate", "deploy"]]
+
+
+def test_migrate_deploy_stops_at_its_own_timeout(
+    toolchain_env: tuple[Path, Path], monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """The deploy budget still bounds a deploy that hangs, so boot cannot wait forever."""
+    _, log_path = toolchain_env
+    monkeypatch.setenv("DATABASE_URL", "postgresql://u:p@localhost:9/x")
+    monkeypatch.setenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, "1")
+    monkeypatch.setenv("FAKE_PRISMA_FIRST_DEPLOY_SLEEP", "60")
+    monkeypatch.setenv("FAKE_PRISMA_LATER_DEPLOY_STDERR", "Error: P3018 permission denied for schema public")
+
+    started = time.monotonic()
+    with pytest.raises(RuntimeError, match="insufficient permissions"):
+        ProxyExtrasDBManager.setup_database(use_migrate=True, use_v2_resolver=True)
+    elapsed = time.monotonic() - started
+
+    assert len(_deploy_calls(log_path)) == 2
+    assert elapsed < 30
+
+
+def test_db_push_timeout_hint_names_the_per_command_budget(
+    toolchain_env: tuple[Path, Path], monkeypatch: pytest.MonkeyPatch, caplog: pytest.LogCaptureFixture
+) -> None:
+    """``db push`` keeps the per-command budget, so its timeout hint has to name that variable."""
+    _, log_path = toolchain_env
+    monkeypatch.setenv("DATABASE_URL", "postgresql://u:p@localhost:9/x")
+    monkeypatch.setenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, "1")
+    monkeypatch.setenv("FAKE_PRISMA_FIRST_PUSH_SLEEP", "3")
+
+    with caplog.at_level(logging.WARNING, logger="litellm_proxy_extras"):
+        assert ProxyExtrasDBManager.setup_database(use_migrate=False, use_v2_resolver=False) is True
+
+    assert [call["args"][:2] for call in _fake_prisma_calls(log_path)].count(["db", "push"]) == 2
+    assert [record.getMessage() for record in caplog.records if "timed out" in record.getMessage()] == [
+        f"Attempt 1 timed out. Raise {PRISMA_COMMAND_TIMEOUT_ENV_VAR} if this database needs longer to apply its schema."
+    ]
+
+
+@pytest.mark.parametrize(
+    ("command_timeout", "deploy_timeout", "expected"),
+    [
+        ("900", None, 900.0),
+        ("12", None, DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT),
+        ("900", "1200", 1200.0),
+        ("900", "300", 300.0),
+    ],
+    ids=["raised_command_budget_carries_over", "lowered_command_budget_does_not", "override_wins_upward", "override_wins_downward"],
+)
+def test_migrate_deploy_budget_keeps_a_raised_command_budget(
+    command_timeout: str, deploy_timeout: str | None, expected: float, monkeypatch: pytest.MonkeyPatch
+) -> None:
+    """Deployments that raised the per-command budget to survive a long deploy keep that budget for deploy."""
+    monkeypatch.setenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, command_timeout)
+    if deploy_timeout is None:
+        monkeypatch.delenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, raising=False)
+    else:
+        monkeypatch.setenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, deploy_timeout)
+
+    assert prisma_migrate_deploy_timeout() == expected
+
+
 @pytest.mark.parametrize(
     "raw",
     ["", "0", "-5", "not-a-number", "nan", "inf", "-inf", "1e400"],
 )
+@pytest.mark.parametrize(
+    ("env_var", "read_timeout", "default"),
+    [
+        (PRISMA_COMMAND_TIMEOUT_ENV_VAR, prisma_command_timeout, DEFAULT_PRISMA_COMMAND_TIMEOUT),
+        (PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, prisma_migrate_deploy_timeout, DEFAULT_PRISMA_MIGRATE_DEPLOY_TIMEOUT),
+    ],
+    ids=["command", "migrate_deploy"],
+)
 def test_unusable_timeout_override_falls_back_to_the_default(
-    raw: str, monkeypatch: pytest.MonkeyPatch
+    raw: str, env_var: str, read_timeout: Callable[[], float], default: float, monkeypatch: pytest.MonkeyPatch
 ) -> None:
     """A non-finite override would silently disable the timeout it configures."""
-    monkeypatch.setenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, raw)
+    monkeypatch.setenv(env_var, raw)
 
-    assert prisma_command_timeout() == DEFAULT_PRISMA_COMMAND_TIMEOUT
+    assert read_timeout() == default
 
 
 def test_timeout_overrides_are_independent(monkeypatch: pytest.MonkeyPatch) -> None:
     monkeypatch.setenv(PRISMA_COMMAND_TIMEOUT_ENV_VAR, "12")
     monkeypatch.setenv(PRISMA_BOOTSTRAP_TIMEOUT_ENV_VAR, "900")
+    monkeypatch.setenv(PRISMA_MIGRATE_DEPLOY_TIMEOUT_ENV_VAR, "1200")
 
     assert prisma_command_timeout() == 12
     assert prisma_bootstrap_timeout() == 900
+    assert prisma_migrate_deploy_timeout() == 1200
 
 
 @pytest.mark.parametrize("module", ["utils.py", "replica_identity.py"])
diff --git a/tests/proxy_unit_tests/test_proxy_server.py b/tests/proxy_unit_tests/test_proxy_server.py
index 47554913419..54cce9cdd78 100644
--- a/tests/proxy_unit_tests/test_proxy_server.py
+++ b/tests/proxy_unit_tests/test_proxy_server.py
@@ -3076,7 +3076,9 @@ async def test_update_config_success_callback_normalization():
     admin_user = UserAPIKeyAuth(
         user_role=LitellmUserRoles.PROXY_ADMIN, api_key="sk-test"
     )
-    await proxy_server.update_config(config_update, user_api_key_dict=admin_user)
+    request = MagicMock()
+    request.json = AsyncMock(return_value={"litellm_settings": {"success_callback": ["SQS", "sQs"]}})
+    await proxy_server.update_config(config_update, request=request, user_api_key_dict=admin_user)
 
     assert (
         "litellm_settings" in upserted
diff --git a/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py b/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py
index 3785ccdcfba..096efc33aaf 100644
--- a/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py
+++ b/tests/proxy_unit_tests/test_unit_test_max_model_budget_limiter.py
@@ -1,3 +1,5 @@
+import asyncio
+from types import MappingProxyType
 from unittest.mock import AsyncMock, patch
 
 
@@ -5,6 +7,7 @@ import pytest
 
 import litellm
 from litellm.caching.caching import DualCache
+from litellm.caching.redis_cache import RedisCache
 from datetime import datetime, timezone
 
 from litellm.litellm_core_utils.duration_parser import duration_in_seconds
@@ -1332,3 +1335,85 @@ async def test_the_user_scope_has_no_pre_upgrade_counter_to_carry():
         await limiter.is_user_within_model_budget(
             user_id="u1", user_model_max_budget=model_max_budget, model="openai/gpt-4"
         )
+
+
+class _SharedFakeRedis(RedisCache):
+    """Stand-in for the one Redis every replica's DualCache is attached to.
+
+    Only the methods the limiter and DualCache call are implemented, and
+    ``super().__init__`` is skipped so no connection is opened.
+    """
+
+    def __init__(self):
+        self._store = MappingProxyType({})
+
+    async def async_set_cache(self, key, value, **kwargs):
+        self._store = MappingProxyType({**self._store, key: value})
+
+    async def async_get_cache(self, key, **kwargs):
+        return self._store.get(key)
+
+    async def async_batch_get_cache(self, key_list, **kwargs):
+        return {key: self._store.get(key) for key in key_list}
+
+    async def async_increment_pipeline(self, increment_list, **kwargs):
+        for op in increment_list:
+            total = self._store.get(op["key"], 0.0) + op["increment_value"]
+            self._store = MappingProxyType({**self._store, op["key"]: total})
+        return [self._store[op["key"]] for op in increment_list]
+
+
+async def _log_spend(limiter, *, key_hash, model_max_budget, response_cost):
+    await limiter.async_log_success_event(
+        _success_kwargs(
+            model_group="gpt-4",
+            response_cost=response_cost,
+            key_hash=key_hash,
+            key_model_max_budget=model_max_budget,
+        ),
+        response_obj=None,
+        start_time=None,
+        end_time=None,
+    )
+    # The Redis push is scheduled as a task rather than awaited inline.
+    await asyncio.gather(*(t for t in asyncio.all_tasks() if t is not asyncio.current_task()))
+
+
+@pytest.mark.asyncio
+async def test_spend_logged_on_one_replica_is_enforced_and_reported_on_another():
+    """
+    Each replica increments its own in-memory copy of the per-model counter and
+    pushes the increment to the shared Redis, so only Redis holds the window's
+    total. A replica that has served part of the traffic must still enforce and
+    report the total, not its own share.
+
+    Regression: reads went to the in-memory tier first, so a replica whose local
+    copy sat under the cap kept admitting requests and /key/info on it reported
+    that local share, while the shared counter was already over the cap.
+    """
+    shared_redis = _SharedFakeRedis()
+    replica_a = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=DualCache(redis_cache=shared_redis))
+    replica_b = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=DualCache(redis_cache=shared_redis))
+    key_hash = "vk-shared"
+    model_max_budget = {"gpt-4": {"budget_limit": 1.0, "time_period": "30d"}}
+    user_api_key = UserAPIKeyAuth(token=key_hash, model_max_budget=model_max_budget)
+
+    await _log_spend(replica_b, key_hash=key_hash, model_max_budget=model_max_budget, response_cost=0.25)
+    await _log_spend(replica_a, key_hash=key_hash, model_max_budget=model_max_budget, response_cost=0.5)
+    await _log_spend(replica_a, key_hash=key_hash, model_max_budget=model_max_budget, response_cost=0.5)
+
+    with pytest.raises(litellm.BudgetExceededError):
+        await replica_b.is_key_within_model_budget(user_api_key, "gpt-4")
+
+    usage_on_b = await build_model_max_budget_usage(
+        entity_type=Litellm_EntityType.KEY,
+        entity_id=key_hash,
+        model_max_budget=model_max_budget,
+        cache=replica_b.dual_cache,
+    )
+    assert usage_on_b["gpt-4"]["current_spend"] == 1.25
+
+    # Control: a replica that never served this key reads the same total.
+    replica_c = _PROXY_VirtualKeyModelMaxBudgetLimiter(dual_cache=DualCache(redis_cache=shared_redis))
+    with pytest.raises(litellm.BudgetExceededError):
+        await replica_c.is_key_within_model_budget(user_api_key, "gpt-4")
diff --git a/tests/router_unit_tests/test_router_endpoints.py b/tests/router_unit_tests/test_router_endpoints.py
index d37af5b456a..a06aaa363ad 100644
--- a/tests/router_unit_tests/test_router_endpoints.py
+++ b/tests/router_unit_tests/test_router_endpoints.py
@@ -1324,6 +1324,98 @@ async def test_init_containers_api_endpoints_managed_id_without_model_id_applies
     assert call_kw["custom_llm_provider"] == "azure"
 
 
+@pytest.mark.asyncio
+async def test_init_containers_api_endpoints_create_with_model_uses_deployment_credentials(monkeypatch):
+    """
+    ``POST /v1/containers`` carries no container ID, so a ``model`` in the request
+    body is the only way to pick a deployment. The upstream call must receive that
+    deployment's ``api_key``/``api_base`` instead of falling back to the global
+    ``OPENAI_API_KEY`` (which may be unset on the proxy).
+    """
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-5.4",
+                "litellm_params": {
+                    "model": "openai/gpt-5.4",
+                    "api_key": "sk-model-list-key",
+                    "api_base": "https://custom.openai.example/v1",
+                },
+            }
+        ]
+    )
+    mock_original_function = AsyncMock(return_value={"id": "cntr_test", "name": "Test Container"})
+
+    await router._init_containers_api_endpoints(
+        original_function=mock_original_function,
+        custom_llm_provider="openai",
+        name="Test Container",
+        model="gpt-5.4",
+    )
+
+    mock_original_function.assert_called_once()
+    call_kw = mock_original_function.call_args.kwargs
+    assert call_kw["api_key"] == "sk-model-list-key"
+    assert call_kw["api_base"] == "https://custom.openai.example/v1"
+    assert call_kw["model"] == "openai/gpt-5.4"
+    assert call_kw["name"] == "Test Container"
+
+
+@pytest.mark.asyncio
+async def test_init_containers_api_endpoints_create_without_model_calls_directly():
+    """
+    Without ``model`` (or with ``model=None`` as the proxy forwards it), create/list
+    must keep calling the handler directly with global provider credentials.
+    """
+    router = Router(model_list=[])
+    router._ageneric_api_call_with_fallbacks = AsyncMock()
+    mock_original_function = AsyncMock(return_value={"id": "cntr_test"})
+
+    await router._init_containers_api_endpoints(
+        original_function=mock_original_function,
+        custom_llm_provider="openai",
+        name="Test Container",
+        model=None,
+    )
+
+    router._ageneric_api_call_with_fallbacks.assert_not_called()
+    mock_original_function.assert_called_once_with(custom_llm_provider="openai", name="Test Container", model=None)
+
+
+@pytest.mark.asyncio
+async def test_init_containers_api_endpoints_create_with_unknown_model_passes_through(monkeypatch):
+    """
+    A ``model`` that names no configured deployment must not turn into a 400. The call
+    falls through to the handler with the caller's model and no injected deployment
+    credentials, matching the behaviour before model-based routing existed.
+    """
+    monkeypatch.delenv("OPENAI_API_KEY", raising=False)
+    router = Router(
+        model_list=[
+            {
+                "model_name": "gpt-5.4",
+                "litellm_params": {"model": "openai/gpt-5.4", "api_key": "sk-model-list-key"},
+            }
+        ]
+    )
+    mock_original_function = AsyncMock(return_value={"id": "cntr_test"})
+
+    await router._init_containers_api_endpoints(
+        original_function=mock_original_function,
+        custom_llm_provider="openai",
+        name="Test Container",
+        model="does-not-exist",
+    )
+
+    mock_original_function.assert_called_once()
+    call_kw = mock_original_function.call_args.kwargs
+    assert call_kw["model"] == "does-not-exist"
+    assert call_kw["name"] == "Test Container"
+    assert "api_key" not in call_kw
+    assert "api_base" not in call_kw
+
+
 def test_router_model_group_encrypted_content_affinity_callback_registration():
     from litellm.router_utils.pre_call_checks.deployment_affinity_check import (
         DeploymentAffinityCheck,
diff --git a/tests/test_litellm/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py b/tests/test_litellm/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py
index 5503a5668bf..a8fe464ec32 100644
--- a/tests/test_litellm/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py
+++ b/tests/test_litellm/a2a_protocol/providers/bedrock_agentcore/test_bedrock_agentcore_a2a.py
@@ -11,7 +11,9 @@ Verifies that:
 
 import json
 
+import httpx
 import pytest
+import respx
 from unittest.mock import AsyncMock, MagicMock, patch
 
 
@@ -295,6 +297,195 @@ class TestTransformation:
         assert headers["Authorization"].startswith("AWS4-HMAC-SHA256")
 
 
+SESSION_HEADER = "X-Amzn-Bedrock-AgentCore-Runtime-Session-Id"
+CONTEXT_ID = "conversation-alpha-0001-0000000000000000"
+KEY_HASH = "hashed-key-of-caller-one"
+
+
+def _params_with_context(context_id: object) -> dict:
+    return {"message": {**SAMPLE_PARAMS["message"], "contextId": context_id}}
+
+
+def _scoped(context_id: str, key_hash: str) -> str:
+    import hashlib
+
+    return f"{hashlib.sha256(key_hash.encode()).hexdigest()[:16]}-{context_id}"
+
+
+def _session_header(params: dict, litellm_params: dict) -> str:
+    from litellm.a2a_protocol.providers.bedrock_agentcore.transformation import (
+        BedrockAgentCoreA2ATransformation,
+    )
+
+    _, headers, _ = BedrockAgentCoreA2ATransformation.get_url_and_signed_request(
+        request_id="req-001",
+        params=params,
+        litellm_params=litellm_params,
+    )
+    return headers[SESSION_HEADER]
+
+
+@pytest.fixture
+def httpx_transport(monkeypatch):
+    import litellm
+
+    monkeypatch.setattr(litellm, "disable_aiohttp_transport", True)
+    monkeypatch.setenv("DISABLE_AIOHTTP_TRANSPORT", "True")
+    litellm.in_memory_llm_clients_cache.flush_cache()
+    yield
+    litellm.in_memory_llm_clients_cache.flush_cache()
+
+
+class TestRequestScopedRuntimeSession:
+    """message.contextId selects the AgentCore runtime session, scoped to the calling key."""
+
+    def test_context_id_scoped_to_calling_key(self):
+        from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+            A2A_USER_API_KEY_HASH_PARAM,
+        )
+
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: KEY_HASH}
+        assert _session_header(_params_with_context(CONTEXT_ID), litellm_params) == _scoped(CONTEXT_ID, KEY_HASH)
+
+    def test_context_id_used_verbatim_without_principal(self):
+        assert _session_header(_params_with_context(CONTEXT_ID), SAMPLE_LITELLM_PARAMS) == CONTEXT_ID
+
+    def test_same_context_id_reuses_session_and_other_context_isolated(self):
+        first = _session_header(_params_with_context(CONTEXT_ID), SAMPLE_LITELLM_PARAMS)
+        second = _session_header(_params_with_context(CONTEXT_ID), SAMPLE_LITELLM_PARAMS)
+        other = _session_header(
+            _params_with_context("conversation-beta-00002-0000000000000000"),
+            SAMPLE_LITELLM_PARAMS,
+        )
+        assert first == second
+        assert other != first
+
+    def test_same_context_id_from_different_keys_is_isolated(self):
+        from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+            A2A_USER_API_KEY_HASH_PARAM,
+        )
+
+        params = _params_with_context(CONTEXT_ID)
+        caller_one = _session_header(params, {**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: KEY_HASH})
+        caller_two = _session_header(
+            params, {**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: "hashed-key-of-caller-two"}
+        )
+        assert caller_one != caller_two
+        assert caller_one.endswith(f"-{CONTEXT_ID}")
+        assert caller_two.endswith(f"-{CONTEXT_ID}")
+
+    def test_context_id_takes_precedence_over_configured_session(self):
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, "runtimeSessionId": "a" * 40}
+        assert _session_header(_params_with_context(CONTEXT_ID), litellm_params) == CONTEXT_ID
+
+    def test_configured_session_is_fallback_without_context_id(self):
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, "runtimeSessionId": "a" * 40}
+        assert _session_header(SAMPLE_PARAMS, litellm_params) == "a" * 40
+        assert _session_header(_params_with_context(""), litellm_params) == "a" * 40
+
+    def test_no_context_id_and_no_config_generates_new_session_per_request(self):
+        first = _session_header(SAMPLE_PARAMS, SAMPLE_LITELLM_PARAMS)
+        second = _session_header(SAMPLE_PARAMS, SAMPLE_LITELLM_PARAMS)
+        assert first != second
+        assert 33 <= len(first) <= 256
+
+    @pytest.mark.parametrize(
+        "context_id",
+        [
+            "short-context-id",
+            "x" * 257,
+        ],
+    )
+    def test_invalid_context_id_rejected_with_clear_error(self, context_id):
+        import litellm
+
+        with pytest.raises(litellm.BadRequestError, match="Invalid AgentCore runtime session id") as exc_info:
+            _session_header(_params_with_context(context_id), SAMPLE_LITELLM_PARAMS)
+        assert exc_info.value.status_code == 400
+        assert "33-256" in str(exc_info.value)
+
+    def test_scoped_context_id_shorter_than_33_rejected(self):
+        import litellm
+        from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+            A2A_USER_API_KEY_HASH_PARAM,
+        )
+
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: KEY_HASH}
+        with pytest.raises(litellm.BadRequestError, match=_scoped("c" * 15, KEY_HASH)):
+            _session_header(_params_with_context("c" * 15), litellm_params)
+        assert _session_header(_params_with_context("c" * 16), litellm_params) == _scoped("c" * 16, KEY_HASH)
+
+    def test_invalid_configured_session_rejected(self):
+        import litellm
+
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, "runtimeSessionId": "too-short"}
+        with pytest.raises(litellm.BadRequestError, match="Invalid AgentCore runtime session id"):
+            _session_header(SAMPLE_PARAMS, litellm_params)
+
+    def test_non_string_context_id_falls_back(self):
+        litellm_params = {**SAMPLE_LITELLM_PARAMS, "runtimeSessionId": "a" * 40}
+        assert _session_header(_params_with_context(12345), litellm_params) == "a" * 40
+
+    def test_spoofed_session_header_does_not_override_context_id(self):
+        from litellm.a2a_protocol.providers.bedrock_agentcore.transformation import (
+            BedrockAgentCoreA2ATransformation,
+        )
+
+        _, headers, _ = BedrockAgentCoreA2ATransformation.get_url_and_signed_request(
+            request_id="req-001",
+            params=_params_with_context(CONTEXT_ID),
+            litellm_params=SAMPLE_LITELLM_PARAMS,
+            agent_extra_headers={SESSION_HEADER: "s" * 40},
+        )
+        assert headers[SESSION_HEADER] == CONTEXT_ID
+
+    @pytest.mark.asyncio
+    async def test_context_id_session_header_on_outbound_non_streaming_post(self, httpx_transport):
+        from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+            A2A_USER_API_KEY_HASH_PARAM,
+        )
+        from litellm.a2a_protocol.providers.bedrock_agentcore.config import (
+            BedrockAgentCoreA2AConfig,
+        )
+
+        with respx.mock(assert_all_called=True) as router:
+            route = router.post(url__regex=r".*/invocations.*").mock(
+                return_value=httpx.Response(200, json={"jsonrpc": "2.0", "id": "req-001", "result": {}})
+            )
+            await BedrockAgentCoreA2AConfig().handle_non_streaming(
+                request_id="req-001",
+                params=_params_with_context(CONTEXT_ID),
+                litellm_params={**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: KEY_HASH},
+            )
+
+        assert route.calls.last.request.headers[SESSION_HEADER] == _scoped(CONTEXT_ID, KEY_HASH)
+
+    @pytest.mark.asyncio
+    async def test_context_id_session_header_on_outbound_streaming_post(self, httpx_transport):
+        from litellm.a2a_protocol.litellm_completion_bridge.handler import (
+            A2A_USER_API_KEY_HASH_PARAM,
+        )
+        from litellm.a2a_protocol.providers.bedrock_agentcore.config import (
+            BedrockAgentCoreA2AConfig,
+        )
+
+        with respx.mock(assert_all_called=True) as router:
+            route = router.post(url__regex=r".*/invocations.*").mock(
+                return_value=httpx.Response(200, json={"jsonrpc": "2.0", "id": "req-001", "result": {}})
+            )
+            events = [
+                event
+                async for event in BedrockAgentCoreA2AConfig().handle_streaming(
+                    request_id="req-001",
+                    params=_params_with_context(CONTEXT_ID),
+                    litellm_params={**SAMPLE_LITELLM_PARAMS, A2A_USER_API_KEY_HASH_PARAM: KEY_HASH},
+                )
+            ]
+
+        assert events == [{"jsonrpc": "2.0", "id": "req-001", "result": {}}]
+        assert route.calls.last.request.headers[SESSION_HEADER] == _scoped(CONTEXT_ID, KEY_HASH)
+
+
 class TestNonStreaming:
     """Test end-to-end non-streaming flow."""
 
diff --git a/tests/test_litellm/containers/test_container_api.py b/tests/test_litellm/containers/test_container_api.py
index 885c4cd294a..16a3844431e 100644
--- a/tests/test_litellm/containers/test_container_api.py
+++ b/tests/test_litellm/containers/test_container_api.py
@@ -152,6 +152,42 @@ class TestContainerAPI:
             assert response.id == "cntr_async_123"
             assert response.name == "Async Test Container"
 
+    @pytest.mark.asyncio
+    async def test_acreate_container_encodes_router_model_id(self):
+        """
+        The async handler returns a coroutine, so the managed-ID encoding must run
+        after it resolves. Otherwise follow-up calls (retrieve/delete/files) lose the
+        deployment and fall back to global provider credentials.
+        """
+        upstream_response = ContainerObject(
+            id="cntr_upstream_123",
+            object="container",
+            created_at=1747857508,
+            status="running",
+            expires_after={"anchor": "last_active_at", "minutes": 20},
+            last_active_at=1747857508,
+            name="Routed Container",
+        )
+
+        async def _resolve_upstream():
+            return upstream_response
+
+        with patch.object(  # test-quality-ok: create_container exposes no client seam, only the handler
+            base_llm_http_handler,
+            "container_create_handler",
+            side_effect=lambda **kwargs: _resolve_upstream() if kwargs["_is_async"] else upstream_response,
+        ):
+            response = await acreate_container(
+                name="Routed Container",
+                custom_llm_provider="openai",
+                litellm_metadata={"model_info": {"id": "deployment-abc"}},
+            )
+
+        decoded = ResponsesAPIRequestUtils._decode_container_id(response.id)
+        assert decoded["model_id"] == "deployment-abc"
+        assert decoded["custom_llm_provider"] == "openai"
+        assert decoded["response_id"] == "cntr_upstream_123"
+
     @pytest.mark.asyncio
     async def test_alist_containers_basic(self):
         """Test basic async container listing functionality."""
diff --git a/tests/test_litellm/integrations/test_custom_guardrail.py b/tests/test_litellm/integrations/test_custom_guardrail.py
index d978eb48c12..7d70b9a8862 100644
--- a/tests/test_litellm/integrations/test_custom_guardrail.py
+++ b/tests/test_litellm/integrations/test_custom_guardrail.py
@@ -2237,3 +2237,202 @@ class TestRecordsOwnGuardrailInformation:
         )
 
         assert _guardrail_entries(request_data) == []
+
+
+class _ApplyOnlyObserver(CustomGuardrail):
+    """Overrides only apply_guardrail, like panw_prisma_airs; inherits async_logging_hook."""
+
+    def __init__(self, block: bool = False):
+        from litellm.types.guardrails import GuardrailEventHooks
+
+        super().__init__(guardrail_name="apply-only-observer", event_hook=GuardrailEventHooks.logging_only)
+        self.block = block
+        self.calls: list = []
+
+    @log_guardrail_information
+    async def apply_guardrail(self, inputs, request_data, input_type, logging_obj=None):
+        from fastapi import HTTPException
+
+        self.calls.append((input_type, list(inputs.get("texts") or [])))
+        if self.block:
+            raise HTTPException(status_code=400, detail={"error": "flagged"})
+        return GenericGuardrailAPIInputs(texts=["[MASKED]" for _ in inputs.get("texts") or []])
+
+
+def _logged_call(messages: list | str) -> tuple[dict, object]:
+    from litellm.types.utils import Choices, Message, ModelResponse
+
+    response = ModelResponse(choices=[Choices(message=Message(role="assistant", content="general kenobi"))])
+    kwargs = {
+        "model": "gpt-5.4-mini",
+        "messages": messages,
+        "litellm_call_id": "call-1",
+        "litellm_params": {"metadata": {"user_api_key_user_id": "u1"}},
+        "optional_params": {},
+        "standard_logging_object": {"guardrail_information": None},
+    }
+    return kwargs, response
+
+
+class TestLoggingOnlyApplyGuardrail:
+    """LIT-4876 regression: a guardrail in mode logging_only that implements only
+    apply_guardrail must still run against the logged request and response and
+    record guardrail_information, instead of inheriting the CustomLogger no-op."""
+
+    @pytest.mark.asyncio
+    async def test_runs_apply_guardrail_observe_only_and_records_verdict(self):
+        guardrail = _ApplyOnlyObserver()
+        messages = [{"role": "user", "content": "hello there"}]
+        kwargs, response = _logged_call(messages)
+
+        out_kwargs, out_response = await guardrail.async_logging_hook(kwargs, response, CallTypes.acompletion.value)
+
+        assert guardrail.calls == [("request", ["hello there"]), ("response", ["general kenobi"])]
+        assert out_kwargs["messages"] == [{"role": "user", "content": "hello there"}]
+        assert out_response.choices[0].message.content == "general kenobi"
+        entries = out_kwargs["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_name"] for e in entries] == ["apply-only-observer", "apply-only-observer"]
+        assert {e["guardrail_mode"] for e in entries} == {"logging_only"}
+        assert {e["guardrail_status"] for e in entries} == {"success"}
+        assert "standard_logging_guardrail_information" not in kwargs["litellm_params"]["metadata"]
+        assert kwargs["standard_logging_object"] == {"guardrail_information": None}
+
+    @pytest.mark.asyncio
+    async def test_appends_to_pre_call_verdicts_without_duplicating_them(self):
+        guardrail = _ApplyOnlyObserver()
+        kwargs, response = _logged_call([{"role": "user", "content": "hello there"}])
+        pre_call_entry = {"guardrail_name": "pii-blocker", "guardrail_mode": "pre_call", "guardrail_status": "success"}
+        kwargs["litellm_params"]["metadata"]["standard_logging_guardrail_information"] = [pre_call_entry]
+        kwargs["standard_logging_object"]["guardrail_information"] = [pre_call_entry]
+
+        out_kwargs, _ = await guardrail.async_logging_hook(kwargs, response, CallTypes.acompletion.value)
+
+        entries = out_kwargs["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_name"] for e in entries] == ["pii-blocker", "apply-only-observer", "apply-only-observer"]
+        assert kwargs["litellm_params"]["metadata"]["standard_logging_guardrail_information"] == [pre_call_entry]
+
+    @pytest.mark.asyncio
+    async def test_request_copy_failure_is_swallowed(self):
+        import threading
+
+        guardrail = _ApplyOnlyObserver()
+        kwargs, response = _logged_call([{"role": "user", "content": "hello there", "lock": threading.Lock()}])
+
+        out_kwargs, out_response = await guardrail.async_logging_hook(kwargs, response, CallTypes.acompletion.value)
+
+        assert guardrail.calls == []
+        assert out_kwargs is kwargs
+        assert out_response is response
+
+    @pytest.mark.asyncio
+    async def test_block_verdict_is_recorded_without_raising(self):
+        guardrail = _ApplyOnlyObserver(block=True)
+        kwargs, response = _logged_call([{"role": "user", "content": "flagged content"}])
+
+        out_kwargs, _ = await guardrail.async_logging_hook(kwargs, response, CallTypes.acompletion.value)
+
+        assert guardrail.calls == [("request", ["flagged content"])]
+        entries = out_kwargs["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_status"] for e in entries] == ["guardrail_intervened"]
+
+    @pytest.mark.asyncio
+    async def test_call_type_without_translation_is_skipped(self):
+        guardrail = _ApplyOnlyObserver()
+        kwargs, response = _logged_call([{"role": "user", "content": "hello there"}])
+
+        out_kwargs, _ = await guardrail.async_logging_hook(kwargs, response, CallTypes.amoderation.value)
+
+        assert guardrail.calls == []
+        assert out_kwargs["standard_logging_object"]["guardrail_information"] is None
+
+    @pytest.mark.asyncio
+    async def test_aembedding_scans_logged_input(self):
+        from litellm.types.utils import EmbeddingResponse
+
+        guardrail = _ApplyOnlyObserver()
+        kwargs, _ = _logged_call("hello there")
+        response = EmbeddingResponse(data=[{"embedding": [0.1], "index": 0, "object": "embedding"}])
+
+        out_kwargs, out_response = await guardrail.async_logging_hook(kwargs, response, CallTypes.aembedding.value)
+
+        assert guardrail.calls == [("request", ["hello there"])]
+        assert out_kwargs["messages"] == "hello there"
+        assert out_response is response
+        entries = out_kwargs["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_status"] for e in entries] == ["success"]
+
+    @pytest.mark.asyncio
+    async def test_native_lifecycle_hook_guardrail_is_left_alone(self):
+        class _NativeHooks(_ApplyOnlyObserver):
+            use_native_lifecycle_hooks = True
+
+        guardrail = _NativeHooks()
+        kwargs, response = _logged_call([{"role": "user", "content": "hello there"}])
+
+        out_kwargs, out_response = await guardrail.async_logging_hook(kwargs, response, CallTypes.acompletion.value)
+
+        assert guardrail.calls == []
+        assert out_kwargs is kwargs
+        assert out_response is response
+
+    @pytest.mark.asyncio
+    async def test_aresponses_scans_logged_messages_when_input_is_cleared(self):
+        from litellm.types.llms.openai import ResponsesAPIResponse
+
+        guardrail = _ApplyOnlyObserver()
+        kwargs, _ = _logged_call([{"role": "user", "content": "hello there"}])
+        kwargs["input"] = None
+        response = ResponsesAPIResponse(
+            id="resp_1",
+            created_at=1,
+            model="gpt-5.4-mini",
+            object="response",
+            status="completed",
+            output=[
+                {
+                    "type": "message",
+                    "id": "msg_1",
+                    "status": "completed",
+                    "role": "assistant",
+                    "content": [{"type": "output_text", "text": "general kenobi"}],
+                }
+            ],
+        )
+
+        out_kwargs, _ = await guardrail.async_logging_hook(kwargs, response, CallTypes.aresponses.value)
+
+        assert guardrail.calls == [("request", ["hello there"]), ("response", ["general kenobi"])]
+        entries = out_kwargs["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_status"] for e in entries] == ["success", "success"]
+
+    @pytest.mark.asyncio
+    async def test_async_success_handler_records_verdict_in_standard_logging_object(self):
+        import datetime as dt
+
+        from litellm.litellm_core_utils.litellm_logging import Logging
+
+        guardrail = _ApplyOnlyObserver()
+        guardrail.default_on = True
+        messages = [{"role": "user", "content": "hello there"}]
+        _, response = _logged_call(messages)
+        logging_obj = Logging(
+            model="gpt-5.4-mini",
+            messages=messages,
+            stream=False,
+            call_type=CallTypes.acompletion.value,
+            start_time=dt.datetime.now(),
+            litellm_call_id="call-1",
+            function_id="fn-1",
+            dynamic_async_success_callbacks=[guardrail],
+        )
+        logging_obj.update_environment_variables(
+            litellm_params={"metadata": {}}, optional_params={}, model="gpt-5.4-mini", custom_llm_provider="openai"
+        )
+
+        await logging_obj.async_success_handler(
+            result=response, start_time=dt.datetime.now(), end_time=dt.datetime.now()
+        )
+
+        assert guardrail.calls == [("request", ["hello there"]), ("response", ["general kenobi"])]
+        entries = logging_obj.model_call_details["standard_logging_object"]["guardrail_information"]
+        assert [e["guardrail_status"] for e in entries] == ["success", "success"]
diff --git a/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py b/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py
index 0e1c832ebf5..b7f0ca1efe1 100644
--- a/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py
+++ b/tests/test_litellm/litellm_core_utils/llm_cost_calc/test_llm_cost_calc_utils.py
@@ -1522,7 +1522,7 @@ def test_gpt_5_6_alias_prices_match_sol(local_model_cost_map):
     sol = litellm.model_cost["gpt-5.6-sol"]
 
     cost_fields = sorted(field for field in sol if "cost" in field)
-    assert len(cost_fields) == 23
+    assert len(cost_fields) == 27
 
     for field in cost_fields:
         assert alias.get(field) == sol.get(field), field
@@ -4039,8 +4039,8 @@ def test_fast_service_tier_matches_priority_above_the_context_threshold(_local_m
     )
 
     assert fast == priority
-    assert fast[0] == pytest.approx(300_000 * 8e-06, rel=1e-9)
-    assert fast[1] == pytest.approx(1_000 * 3e-05, rel=1e-9)
+    assert fast[0] == pytest.approx(300_000 * 1.6e-05, rel=1e-9)
+    assert fast[1] == pytest.approx(1_000 * 6e-05, rel=1e-9)
 
 
 def test_priority_reasoning_tokens_bill_at_the_priority_output_rate(_local_model_cost_map):
@@ -4200,6 +4200,86 @@ def test_generic_cost_per_token_gemini_37_flash(_local_model_cost_map):
     assert completion_cost == pytest.approx(0.001875)
 
 
+GEMINI_38_FLASH_LAUNCH_PRICING = [
+    ("gemini-3.8-flash", 7.5e-07, 3.75e-06, 7.5e-08),
+    ("gemini/gemini-3.8-flash", 7.5e-07, 3.75e-06, 7.5e-08),
+    ("vertex_ai/gemini-3.8-flash", 7.5e-07, 3.75e-06, 7.5e-08),
+]
+
+
+@pytest.mark.parametrize("model,input_cost,output_cost,cache_read_cost", GEMINI_38_FLASH_LAUNCH_PRICING)
+def test_gemini_38_flash_launch_pricing(model, input_cost, output_cost, cache_read_cost, _local_model_cost_map):
+    model_cost_map = litellm.model_cost[model]
+    assert model_cost_map["input_cost_per_token"] == input_cost
+    assert model_cost_map["output_cost_per_token"] == output_cost
+    assert model_cost_map["output_cost_per_reasoning_token"] == output_cost
+    assert model_cost_map["cache_read_input_token_cost"] == cache_read_cost
+    assert model_cost_map["mode"] == "chat"
+    assert model_cost_map["supports_reasoning"] is True
+    assert model_cost_map["supports_function_calling"] is True
+    assert model_cost_map["max_input_tokens"] == 1048576
+
+
+GEMINI_38_FLASH_FIELDS_SHARED_WITH_37_FLASH = (
+    "input_cost_per_token",
+    "output_cost_per_token",
+    "output_cost_per_reasoning_token",
+    "cache_read_input_token_cost",
+    "input_cost_per_token_batches",
+    "output_cost_per_token_batches",
+    "input_cost_per_token_flex",
+    "output_cost_per_token_flex",
+    "cache_read_input_token_cost_flex",
+    "input_cost_per_token_priority",
+    "output_cost_per_token_priority",
+    "cache_read_input_token_cost_priority",
+    "search_context_cost_per_query",
+    "google_maps_grounding_cost_per_query",
+    "prompt_cache_min_tokens",
+    "max_input_tokens",
+    "max_output_tokens",
+    "supports_reasoning",
+    "supports_function_calling",
+    "supports_prompt_caching",
+    "supports_vision",
+    "supports_pdf_input",
+    "supports_audio_input",
+    "supports_video_input",
+    "supports_response_schema",
+    "supports_tool_choice",
+    "supports_web_search",
+    "supports_url_context",
+)
+
+
+@pytest.mark.parametrize("prefix", ["", "gemini/", "vertex_ai/"])
+def test_gemini_38_flash_matches_37_flash_promotional_pricing(prefix, _local_model_cost_map):
+    new_model = litellm.model_cost[f"{prefix}gemini-3.8-flash"]
+    old_model = litellm.model_cost[f"{prefix}gemini-3.7-flash"]
+    for field in GEMINI_38_FLASH_FIELDS_SHARED_WITH_37_FLASH:
+        assert new_model[field] == old_model[field], field
+
+
+def test_generic_cost_per_token_gemini_38_flash(_local_model_cost_map):
+    usage = Usage(
+        prompt_tokens=1000,
+        completion_tokens=500,
+        total_tokens=1500,
+        completion_tokens_details=CompletionTokensDetailsWrapper(
+            reasoning_tokens=200,
+            text_tokens=300,
+        ),
+        prompt_tokens_details=PromptTokensDetailsWrapper(text_tokens=1000),
+    )
+    prompt_cost, completion_cost = generic_cost_per_token(
+        model="gemini-3.8-flash",
+        usage=usage,
+        custom_llm_provider="gemini",
+    )
+    assert prompt_cost == pytest.approx(0.00075)
+    assert completion_cost == pytest.approx(0.001875)
+
+
 def test_grok_46_launch_pricing(_local_model_cost_map):
     model_cost_map = litellm.model_cost["xai/grok-4.6"]
     assert model_cost_map["input_cost_per_token"] == 2e-06
diff --git a/tests/test_litellm/litellm_core_utils/test_streaming_chunk_builder_utils.py b/tests/test_litellm/litellm_core_utils/test_streaming_chunk_builder_utils.py
index 8ac050a04f9..bacbcbf132b 100644
--- a/tests/test_litellm/litellm_core_utils/test_streaming_chunk_builder_utils.py
+++ b/tests/test_litellm/litellm_core_utils/test_streaming_chunk_builder_utils.py
@@ -592,6 +592,59 @@ def test_stream_chunk_builder_litellm_usage_chunks():
     assert usage.total_tokens == 77
 
 
+def test_calculate_usage_honors_openai_sdk_completion_usage_chunks():
+    from openai.types.completion_usage import CompletionUsage
+
+    content_chunk = ModelResponseStream(
+        id="chatcmpl-sdk-usage-1",
+        created=1745513206,
+        model="mantle-claude",
+        object="chat.completion.chunk",
+        system_fingerprint=None,
+        choices=[
+            StreamingChoices(
+                finish_reason="stop",
+                index=0,
+                delta=Delta(
+                    provider_specific_fields=None,
+                    content="ok",
+                    role=None,
+                    function_call=None,
+                    tool_calls=None,
+                    audio=None,
+                ),
+                logprobs=None,
+            )
+        ],
+        provider_specific_fields=None,
+        stream_options={"include_usage": True},
+    )
+    usage_chunk = ModelResponseStream(
+        id="chatcmpl-sdk-usage-1",
+        created=1745513207,
+        model="mantle-claude",
+        object="chat.completion.chunk",
+        system_fingerprint=None,
+        choices=[],
+        provider_specific_fields=None,
+        stream_options={"include_usage": True},
+    )
+    usage_chunk.usage = CompletionUsage(
+        prompt_tokens=20, completion_tokens=60, total_tokens=80, cost=0.000704
+    )
+    assert type(usage_chunk.usage) is CompletionUsage
+
+    chunks = [content_chunk, usage_chunk]
+    usage = ChunkProcessor(chunks=chunks).calculate_usage(
+        chunks=chunks, model="mantle-claude", completion_output=""
+    )
+
+    assert usage.prompt_tokens == 20
+    assert usage.completion_tokens == 60
+    assert usage.total_tokens == 80
+    assert getattr(usage, "cost", None) == pytest.approx(0.000704)
+
+
 def test_get_model_from_chunks_azure_model_router():
     """
     Test that _get_model_from_chunks finds the actual model from Azure Model Router chunks.
diff --git a/tests/test_litellm/llms/bedrock/chat/test_bedrock_converse_handler.py b/tests/test_litellm/llms/bedrock/chat/test_bedrock_converse_handler.py
index 8e67a7e3438..21e3239f623 100644
--- a/tests/test_litellm/llms/bedrock/chat/test_bedrock_converse_handler.py
+++ b/tests/test_litellm/llms/bedrock/chat/test_bedrock_converse_handler.py
@@ -96,10 +96,8 @@ def _completion_kwargs(**overrides):
     return kwargs
 
 
-def _run(**overrides):
-    with patch.object(
-        BedrockConverseLLM, "get_credentials", return_value=RESOLVED_CREDENTIALS
-    ):
+def _run(*, credentials: Credentials | None = RESOLVED_CREDENTIALS, **overrides):
+    with patch.object(BedrockConverseLLM, "get_credentials", return_value=credentials):
         return BedrockConverseLLM().completion(**_completion_kwargs(**overrides))
 
 
@@ -360,7 +358,7 @@ async def test_async_completion_logs_pre_call_by_default():
 
 def _sync_client_returning_converse_response():
     client = MagicMock()
-    client.post = lambda **_kwargs: httpx.Response(
+    client.post.side_effect = lambda **_kwargs: httpx.Response(
         200,
         json=CONVERSE_RESPONSE,
         request=httpx.Request("POST", "https://bedrock-runtime.us-west-2.amazonaws.com"),
@@ -487,3 +485,31 @@ def test_post_call_is_not_logged_twice_when_the_sync_rust_call_declines():
     assert response.choices[0].message.content == "hi"
     assert len(calls["post_call"]) == 1
     assert "hi" in calls["post_call"][0]["original_response"]
+
+
+def test_bearer_token_auth_serves_when_boto3_resolves_no_sigv4_credentials(monkeypatch):
+    """With only `AWS_BEARER_TOKEN_BEDROCK` configured boto3 resolves no
+    credentials at all. Preparing the Rust handoff must not dereference that
+    None: the bearer token signs the request on its own."""
+    monkeypatch.setenv("AWS_BEARER_TOKEN_BEDROCK", "bedrock-bearer-token")
+    client = _sync_client_returning_converse_response()
+
+    response = _run(credentials=None, litellm_params={}, client=client)
+
+    assert response.choices[0].message.content == "hi"
+    sent_headers = client.post.call_args.kwargs["headers"]
+    assert sent_headers["Authorization"] == "Bearer bedrock-bearer-token"
+
+
+def test_the_rust_opt_in_needs_no_sigv4_principal():
+    """The core resolves the bearer token itself, so a bearer-only deployment
+    keeps its opt-in and the gate sees no aws_* credential keys to sign with."""
+    seen = _inject()
+
+    response = _run(credentials=None, api_key="bedrock-bearer-token")
+
+    assert response.choices[0].message.content == "hello from rust"
+    params = seen["call"][0]["optional_params"]
+    assert not {"aws_access_key_id", "aws_secret_access_key", "aws_session_token"} & params.keys()
+    assert params["aws_region_name"] == "us-east-1"
+    assert seen["call"][0]["api_key"] == "bedrock-bearer-token"
diff --git a/tests/test_litellm/llms/bedrock/test_base_aws_llm.py b/tests/test_litellm/llms/bedrock/test_base_aws_llm.py
index 7d07ac947b1..f854d806bdc 100644
--- a/tests/test_litellm/llms/bedrock/test_base_aws_llm.py
+++ b/tests/test_litellm/llms/bedrock/test_base_aws_llm.py
@@ -15,6 +15,7 @@ from unittest.mock import MagicMock, patch
 from botocore.awsrequest import AWSPreparedRequest, AWSRequest
 from botocore.auth import SigV4Auth
 from botocore.credentials import Credentials
+from botocore.exceptions import NoCredentialsError
 
 import litellm
 from litellm.llms.bedrock.base_aws_llm import (
@@ -801,6 +802,23 @@ def test_get_request_headers_with_sigv4():
         assert result == mock_request.prepare.return_value
 
 
+def test_get_request_headers_without_credentials_or_bearer_token_raises_no_credentials():
+    """Bearer-token auth needs no SigV4 principal, so `credentials` may be None.
+    Reaching the SigV4 branch with neither must fail the way botocore always
+    has instead of signing with a missing principal."""
+    llm = BaseAWSLLM()
+
+    with patch.dict(os.environ, {}, clear=True), pytest.raises(NoCredentialsError):
+        llm.get_request_headers(
+            credentials=None,
+            aws_region_name="us-west-2",
+            extra_headers=None,
+            endpoint_url="https://api.example.com",
+            data='{"prompt": "test"}',
+            headers={"Content-Type": "application/json"},
+        )
+
+
 def test_sigv4_matches_rust_golden_vector():
     request = AWSRequest(
         method="POST",
diff --git a/tests/test_litellm/llms/s3_vectors/vector_stores/test_s3_vectors_transformation.py b/tests/test_litellm/llms/s3_vectors/vector_stores/test_s3_vectors_transformation.py
index 7085e45cdc3..4b58d220623 100644
--- a/tests/test_litellm/llms/s3_vectors/vector_stores/test_s3_vectors_transformation.py
+++ b/tests/test_litellm/llms/s3_vectors/vector_stores/test_s3_vectors_transformation.py
@@ -1,4 +1,4 @@
-from unittest.mock import MagicMock, Mock
+from unittest.mock import AsyncMock, MagicMock, Mock, patch
 
 import httpx
 import pytest
@@ -9,6 +9,18 @@ from litellm.llms.s3_vectors.vector_stores.transformation import (
 from litellm.types.vector_stores import VectorStoreSearchResponse
 
 
+def _mock_router(model_names, sync=False):
+    """Router mock serving the given embedding model names."""
+    router = MagicMock()
+    router.get_model_list.return_value = [{"model_name": name} for name in model_names]
+    embedding_response = Mock(data=[{"embedding": [0.1, 0.2, 0.3]}])
+    if sync:
+        router.embedding = MagicMock(return_value=embedding_response)
+    else:
+        router.aembedding = AsyncMock(return_value=embedding_response)
+    return router
+
+
 class TestS3VectorsVectorStoreConfig:
     def test_init(self):
         """Test that S3VectorsVectorStoreConfig initializes correctly"""
@@ -28,19 +40,174 @@ class TestS3VectorsVectorStoreConfig:
         url = config.get_complete_url(None, litellm_params)
         assert url == "https://s3vectors.us-west-2.api.aws"
 
-    def test_get_complete_url_missing_region(self):
-        """Test that missing region raises error"""
+    def test_get_complete_url_missing_region(self, monkeypatch):
+        """Missing region falls back to the default region (parity with ingestion)"""
+        monkeypatch.delenv("AWS_REGION_NAME", raising=False)
+        monkeypatch.delenv("AWS_REGION", raising=False)
         config = S3VectorsVectorStoreConfig()
-        litellm_params = {}
-        with pytest.raises(ValueError, match="aws_region_name is required"):
-            config.get_complete_url(None, litellm_params)
+        url = config.get_complete_url(None, {})
+        assert url == "https://s3vectors.us-west-2.api.aws"
+
+    def test_get_complete_url_uses_env_region(self, monkeypatch):
+        """Missing region param resolves from AWS_REGION_NAME env var"""
+        monkeypatch.setenv("AWS_REGION_NAME", "eu-west-1")
+        monkeypatch.delenv("AWS_REGION", raising=False)
+        config = S3VectorsVectorStoreConfig()
+        url = config.get_complete_url(None, {})
+        assert url == "https://s3vectors.eu-west-1.api.aws"
+
+    def test_get_complete_url_invalid_region_format(self):
+        """Invalid region format raises"""
+        config = S3VectorsVectorStoreConfig()
+        with pytest.raises(ValueError, match="Invalid AWS region format"):
+            config.get_complete_url(None, {"aws_region_name": "Bad_Region!"})
 
-    @pytest.mark.skip(reason="Requires embedding API call, tested in integration tests")
     def test_transform_search_request(self):
-        """Test search request transformation"""
-        # This test requires making an actual embedding API call
-        # It's better tested in integration tests
-        pass
+        """Full request-body transformation with a router-injected embedding"""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+        router = _mock_router(["text-embedding-3-small"], sync=True)
+
+        url, request_body = config.transform_search_vector_store_request(
+            vector_store_id="test-bucket:test-index",
+            query="test query",
+            vector_store_search_optional_params={"max_num_results": 7},
+            api_base="https://s3vectors.us-west-2.api.aws",
+            litellm_logging_obj=mock_logging_obj,
+            litellm_params={},
+            extra_body=None,
+            router=router,
+        )
+
+        assert url == "https://s3vectors.us-west-2.api.aws/QueryVectors"
+        assert request_body == {
+            "vectorBucketName": "test-bucket",
+            "indexName": "test-index",
+            "queryVector": {"float32": [0.1, 0.2, 0.3]},
+            "topK": 7,
+            "returnDistance": True,
+            "returnMetadata": True,
+        }
+        assert mock_logging_obj.model_call_details["query"] == "test query"
+
+    @pytest.mark.asyncio
+    async def test_atransform_search_uses_router_for_virtual_model(self):
+        """Regression: router-served embedding models must resolve via the router,
+        not a bare litellm.aembedding call (which has no deployment credentials)."""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+        router = _mock_router(["my-embedding-model"])
+
+        with patch("litellm.aembedding", new=AsyncMock()) as mock_bare_aembedding:  # test-quality-ok: guards that the bare-embedding path is not taken; dispatch seam is the behavior under test
+            url, request_body = await config.atransform_search_vector_store_request(
+                vector_store_id="test-bucket:test-index",
+                query="test query",
+                vector_store_search_optional_params={},
+                api_base="https://s3vectors.us-west-2.api.aws",
+                litellm_logging_obj=mock_logging_obj,
+                litellm_params={"embedding_model": "my-embedding-model"},
+                extra_body=None,
+                router=router,
+            )
+
+        router.aembedding.assert_awaited_once_with(model="my-embedding-model", input=["test query"])
+        mock_bare_aembedding.assert_not_awaited()
+        assert request_body["queryVector"]["float32"] == [0.1, 0.2, 0.3]
+        assert request_body["topK"] == 5  # default
+
+    @pytest.mark.asyncio
+    async def test_atransform_search_falls_back_when_router_does_not_serve_model(self):
+        """Router present but embedding_model is not a router deployment ->
+        bare litellm.aembedding keeps working (provider-prefixed + env creds stores)."""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+        router = _mock_router(["some-other-model"])
+
+        mock_bare = AsyncMock(return_value=Mock(data=[{"embedding": [0.4, 0.5]}]))
+        with patch("litellm.aembedding", new=mock_bare):  # test-quality-ok: stubs the bare-embedding fallback whose request body the test asserts on
+            _, request_body = await config.atransform_search_vector_store_request(
+                vector_store_id="test-bucket:test-index",
+                query="test query",
+                vector_store_search_optional_params={},
+                api_base="https://s3vectors.us-west-2.api.aws",
+                litellm_logging_obj=mock_logging_obj,
+                litellm_params={"embedding_model": "azure/text-embedding-3-small"},
+                extra_body=None,
+                router=router,
+            )
+
+        mock_bare.assert_awaited_once_with(model="azure/text-embedding-3-small", input=["test query"])
+        router.aembedding.assert_not_awaited()
+        assert request_body["queryVector"]["float32"] == [0.4, 0.5]
+
+    @pytest.mark.asyncio
+    async def test_atransform_search_without_router_uses_bare_embedding(self):
+        """Backward compat: no router -> bare litellm.aembedding as before"""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+
+        mock_bare = AsyncMock(return_value=Mock(data=[{"embedding": [0.6, 0.7]}]))
+        with patch("litellm.aembedding", new=mock_bare):  # test-quality-ok: stubs the bare-embedding fallback whose request body the test asserts on
+            _, request_body = await config.atransform_search_vector_store_request(
+                vector_store_id="test-bucket:test-index",
+                query="test query",
+                vector_store_search_optional_params={},
+                api_base="https://s3vectors.us-west-2.api.aws",
+                litellm_logging_obj=mock_logging_obj,
+                litellm_params={},
+                extra_body=None,
+            )
+
+        mock_bare.assert_awaited_once_with(model="text-embedding-3-small", input=["test query"])
+        assert request_body["queryVector"]["float32"] == [0.6, 0.7]
+
+    def test_transform_search_uses_router_for_virtual_model_sync(self):
+        """Sync twin: router-served embedding model resolves via router.embedding"""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+        router = _mock_router(["my-embedding-model"], sync=True)
+
+        with patch("litellm.embedding", new=MagicMock()) as mock_bare_embedding:  # test-quality-ok: guards that the bare-embedding path is not taken; dispatch seam is the behavior under test
+            _, request_body = config.transform_search_vector_store_request(
+                vector_store_id="test-bucket:test-index",
+                query="test query",
+                vector_store_search_optional_params={},
+                api_base="https://s3vectors.us-west-2.api.aws",
+                litellm_logging_obj=mock_logging_obj,
+                litellm_params={"embedding_model": "my-embedding-model"},
+                extra_body=None,
+                router=router,
+            )
+
+        router.embedding.assert_called_once_with(model="my-embedding-model", input=["test query"])
+        mock_bare_embedding.assert_not_called()
+        assert request_body["queryVector"]["float32"] == [0.1, 0.2, 0.3]
+
+    def test_transform_search_without_router_uses_bare_embedding_sync(self):
+        """Sync twin: no router -> bare litellm.embedding as before"""
+        config = S3VectorsVectorStoreConfig()
+        mock_logging_obj = Mock()
+        mock_logging_obj.model_call_details = {}
+
+        mock_bare = MagicMock(return_value=Mock(data=[{"embedding": [0.8, 0.9]}]))
+        with patch("litellm.embedding", new=mock_bare):  # test-quality-ok: stubs the bare-embedding fallback whose request body the test asserts on
+            _, request_body = config.transform_search_vector_store_request(
+                vector_store_id="test-bucket:test-index",
+                query="test query",
+                vector_store_search_optional_params={},
+                api_base="https://s3vectors.us-west-2.api.aws",
+                litellm_logging_obj=mock_logging_obj,
+                litellm_params={},
+                extra_body=None,
+            )
+
+        mock_bare.assert_called_once_with(model="text-embedding-3-small", input=["test query"])
+        assert request_body["queryVector"]["float32"] == [0.8, 0.9]
 
     def test_transform_search_request_invalid_vector_store_id(self):
         """Test that invalid vector_store_id format raises error"""
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
index 8c1de12e7d9..4679b978f78 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_ai_gemini_transformation.py
@@ -1096,10 +1096,13 @@ def test_natively_signed_parallel_turn_never_carries_a_placeholder(model):
         "gemini-3.5-flash",
         "gemini-3.6-flash",
         "gemini-3.7-flash",
+        "gemini-3.8-flash",
         "vertex_ai/gemini-3.5-flash",
         "vertex_ai/gemini-3.7-flash",
+        "vertex_ai/gemini-3.8-flash",
         "gemini/gemini-3.5-flash",
         "gemini/gemini-3.7-flash",
+        "gemini/gemini-3.8-flash",
     ],
 )
 def test_placeholder_scoped_to_first_call_across_gemini_3_variants(model):
diff --git a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
index bd07bec900f..d2788408e09 100644
--- a/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
+++ b/tests/test_litellm/llms/vertex_ai/gemini/test_vertex_and_google_ai_studio_gemini.py
@@ -1185,6 +1185,18 @@ def test_vertex_ai_map_thinking_param_with_budget_tokens_0():
     }
 
 
+def test_vertex_ai_map_thinking_param_without_budget_tokens_for_gemini_3():
+    v = VertexGeminiConfig()
+    result = v.map_openai_params(
+        non_default_params={"thinking": {"type": "enabled"}},
+        optional_params={},
+        model="gemini-3.5-flash",
+        drop_params=False,
+    )
+
+    assert result["thinkingConfig"] == {"includeThoughts": True}
+
+
 def test_vertex_ai_map_tools():
     v = VertexGeminiConfig()
     optional_params = {}
diff --git a/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py b/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
index 2ff38af80b1..43034f889f6 100644
--- a/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
+++ b/tests/test_litellm/proxy/agent_endpoints/test_a2a_endpoints.py
@@ -918,6 +918,62 @@ async def test_task_method_failure_hook_uses_enriched_request_data():
     assert failure_data.get("agent_id") == "test-agent"
 
 
+@pytest.mark.asyncio
+async def test_agentcore_invalid_context_id_returns_jsonrpc_invalid_params_400():
+    from litellm.proxy._types import UserAPIKeyAuth
+
+    agent = _make_agent_mock()
+    agent.litellm_params = {
+        "custom_llm_provider": "bedrock",
+        "model": "bedrock/agentcore/arn:aws:bedrock-agentcore:us-west-2:123456789012:runtime/demo",
+        "api_key": "test-jwt-token",
+    }
+    mock_request = _make_request_mock(
+        "message/send",
+        {
+            "message": {
+                "role": "user",
+                "parts": [{"kind": "text", "text": "Hello"}],
+                "messageId": "msg-1",
+                "contextId": "too-short",
+            }
+        },
+    )
+    user_api_key_dict = UserAPIKeyAuth(api_key="sk-test", user_id="u1", team_id="t1")
+
+    mock_proxy_logging = MagicMock()
+    mock_proxy_logging.pre_call_hook = AsyncMock(
+        side_effect=lambda user_api_key_dict, data, call_type: data
+    )
+    mock_proxy_logging.post_call_failure_hook = AsyncMock(return_value=None)
+
+    with ExitStack() as stack:
+        for p in _base_patches(agent):
+            stack.enter_context(p)
+        stack.enter_context(
+            patch(  # test-quality-ok: same proxy_logging_obj injection the sibling failure-hook test uses; no HTTP call is made because the request is rejected before signing
+                "litellm.proxy.proxy_server.proxy_logging_obj", mock_proxy_logging
+            )
+        )
+
+        from litellm.proxy.agent_endpoints.a2a_endpoints import invoke_agent_a2a
+
+        response = await invoke_agent_a2a(
+            agent_id="test-agent",
+            request=mock_request,
+            fastapi_response=MagicMock(),
+            user_api_key_dict=user_api_key_dict,
+        )
+
+    body = json.loads(response.body.decode())
+    assert response.status_code == 400
+    assert body["id"] == "req-1"
+    assert body["error"]["code"] == -32602
+    assert "Invalid AgentCore runtime session id" in body["error"]["message"]
+    assert "Internal error" not in body["error"]["message"]
+    mock_proxy_logging.post_call_failure_hook.assert_awaited_once()
+
+
 @pytest.mark.asyncio
 async def test_get_extended_agent_card_rewrites_url():
     from litellm.proxy._types import UserAPIKeyAuth
diff --git a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
index 1fbc975e40a..c04fb7b30ec 100644
--- a/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
+++ b/tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py
@@ -2199,6 +2199,164 @@ async def test_history_is_still_compressed(guardrail: HeadroomGuardrail):
     assert has_headroom_retrieve_tool(result.get("tools") or [])
 
 
+# ---------------------------------------------------------------------------
+# #38558: a client that runs its own tool loop (e.g. Claude Code via the MCP
+# gateway) executes headroom_retrieve and echoes the recovered original content
+# back as a tool result. Compressing that row re-derives the same content hash
+# it was just retrieved from -- the marker returns and the agent loops. The
+# retrieved row must be held back from the compression service.
+# ---------------------------------------------------------------------------
+
+RETRIEVE_ECHO_MESSAGES = [
+    {"role": "system", "content": "You are Claude Code. " + "S" * 5000},
+    {"role": "user", "content": "H" * 5000},
+    {
+        "role": "assistant",
+        "content": "Expanding the marker.",
+        "tool_calls": [
+            {
+                "id": "hr_1",
+                "type": "function",
+                "function": {
+                    "name": "mcp__headroom__headroom_retrieve",
+                    "arguments": '{"hash": "b573993006976af767214fac"}',
+                },
+            }
+        ],
+    },
+    {"role": "tool", "tool_call_id": "hr_1", "content": "RETRIEVED BODY " + "R" * 5000},
+    {"role": "assistant", "content": "Older answer. " + "O" * 5000},
+    {"role": "user", "content": "now summarize the description"},
+]
+
+
+@pytest.mark.asyncio
+async def test_retrieved_content_is_never_recompressed(guardrail: HeadroomGuardrail):
+    """The tool result carrying headroom_retrieve output is held back, so it can
+    never collapse back to the hash it was just retrieved from."""
+    wire, result = await _wire_and_result(guardrail, RETRIEVE_ECHO_MESSAGES)
+
+    assert not any(row.get("tool_call_id") == "hr_1" for row in wire)
+    assert not any("RETRIEVED BODY" in json.dumps(row) for row in wire)
+    # Reaches the model byte-identical, so no marker stands in for the expansion.
+    assert result["structured_messages"][3] == RETRIEVE_ECHO_MESSAGES[3]
+    # Negative control: unrelated history is still compressed, not a no-op.
+    assert any(row.get("content") == "H" * 5000 for row in wire)
+
+
+@pytest.mark.asyncio
+async def test_retrieved_content_guard_matches_direct_tool_name(guardrail: HeadroomGuardrail):
+    """Server-side the tool is named headroom_retrieve (no MCP prefix); its
+    result must be protected the same way."""
+    messages = [
+        {"role": "system", "content": "sys " + "S" * 5000},
+        {"role": "user", "content": "H" * 5000},
+        {
+            "role": "assistant",
+            "content": "",
+            "tool_calls": [
+                {
+                    "id": "hr_direct",
+                    "type": "function",
+                    "function": {"name": HEADROOM_RETRIEVE_TOOL_NAME, "arguments": "{}"},
+                }
+            ],
+        },
+        {"role": "tool", "tool_call_id": "hr_direct", "content": "RETRIEVED BODY " + "R" * 5000},
+        {"role": "assistant", "content": "Older. " + "O" * 5000},
+        {"role": "user", "content": "summarize"},
+    ]
+    wire, result = await _wire_and_result(guardrail, messages)
+
+    assert not any(row.get("tool_call_id") == "hr_direct" for row in wire)
+    assert result["structured_messages"][3] == messages[3]
+
+
+@pytest.mark.asyncio
+async def test_retrieved_content_protected_when_mcp_tool_name_is_truncated(guardrail: HeadroomGuardrail):
+    """A long mcp__<server>__headroom_retrieve name is truncated past 64 chars in
+    the OpenAI-translated view the guardrail scans, dropping the suffix. The call
+    id read from the request's own Anthropic tool_use (never truncated) still
+    pairs the retrieved row so it is held back."""
+    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        truncate_tool_name,
+    )
+
+    long_name = "mcp__" + "s" * 45 + "__" + HEADROOM_RETRIEVE_TOOL_NAME
+    assert len(long_name) > 64
+    truncated = truncate_tool_name(long_name)
+    assert not truncated.endswith(HEADROOM_RETRIEVE_TOOL_NAME)
+
+    # What the guardrail scans: OpenAI-translated messages with the truncated name.
+    structured = [
+        {"role": "system", "content": "sys " + "S" * 5000},
+        {"role": "user", "content": "H" * 5000},
+        {
+            "role": "assistant",
+            "content": "",
+            "tool_calls": [{"id": "hr_long", "type": "function", "function": {"name": truncated, "arguments": "{}"}}],
+        },
+        {"role": "tool", "tool_call_id": "hr_long", "content": "RETRIEVED BODY " + "R" * 5000},
+        {"role": "assistant", "content": "Older. " + "O" * 5000},
+        {"role": "user", "content": "summarize"},
+    ]
+    # The request's own messages, untranslated: Anthropic tool_use carries the full name.
+    raw_messages = [
+        {"role": "assistant", "content": [{"type": "tool_use", "id": "hr_long", "name": long_name, "input": {}}]},
+        {"role": "user", "content": [{"type": "tool_result", "tool_use_id": "hr_long", "content": "RETRIEVED BODY"}]},
+    ]
+
+    inputs = GenericGuardrailAPIInputs(texts=["x"], structured_messages=json.loads(json.dumps(structured)))
+    sent: dict = {}
+
+    def _echo(**kwargs):
+        sent["messages"] = kwargs["json"]["messages"]
+        return _make_compress_response(json.loads(json.dumps(kwargs["json"]["messages"])))
+
+    with patch.object(guardrail.async_handler, "post", new_callable=AsyncMock, side_effect=_echo):
+        result = await guardrail.apply_guardrail(
+            inputs=inputs,
+            request_data={"model": "claude-sonnet-4-5-20250929", "messages": raw_messages},
+            input_type="request",
+        )
+
+    assert not any(row.get("tool_call_id") == "hr_long" for row in sent["messages"])
+    assert result["structured_messages"][3] == structured[3]
+    assert any(row.get("content") == "H" * 5000 for row in sent["messages"])
+
+
+def test_raw_retrieve_call_ids_covers_both_shapes_and_ignores_others():
+    """Retrieve ids are read from OpenAI tool_calls and Anthropic tool_use blocks;
+    non-retrieve calls, non-tool_use blocks, string content, and non-list inputs
+    yield nothing."""
+    from litellm.proxy.guardrails.guardrail_hooks.headroom.headroom import _raw_retrieve_call_ids
+
+    messages = [
+        {
+            "role": "assistant",
+            "tool_calls": [
+                {"id": "oa1", "function": {"name": HEADROOM_RETRIEVE_TOOL_NAME}},
+                {"id": "other", "function": {"name": "get_weather"}},
+                {"id": "malformed", "function": {"name": 123}},
+                {"id": "nofunc"},
+            ],
+        },
+        {
+            "role": "assistant",
+            "content": [
+                {"type": "tool_use", "id": "an1", "name": "mcp__hr__headroom_retrieve", "input": {}},
+                {"type": "tool_use", "id": "an2", "name": "jira_get_issue", "input": {}},
+                {"type": "text", "text": "noise"},
+            ],
+        },
+        {"role": "user", "content": "plain string content, not a list"},
+    ]
+
+    assert _raw_retrieve_call_ids(messages) == frozenset({"oa1", "an1"})
+    assert _raw_retrieve_call_ids("not a list") == frozenset()
+    assert _raw_retrieve_call_ids(None) == frozenset()
+
+
 @pytest.mark.asyncio
 async def test_nothing_compressible_returns_inputs_untouched(guardrail: HeadroomGuardrail):
     """A single-turn request is all protected, so there is nothing to send and
diff --git a/tests/test_litellm/proxy/management_helpers/test_access_group_key_sync.py b/tests/test_litellm/proxy/management_helpers/test_access_group_key_sync.py
new file mode 100644
index 00000000000..60c36e33e09
--- /dev/null
+++ b/tests/test_litellm/proxy/management_helpers/test_access_group_key_sync.py
@@ -0,0 +1,57 @@
+from types import SimpleNamespace
+from unittest.mock import AsyncMock, MagicMock
+
+import pytest
+
+from litellm.proxy.db.prisma_client import PrismaWrapper
+from litellm.proxy.db.routing_prisma_wrapper import RoutingPrismaWrapper
+from litellm.proxy.management_helpers.access_group_key_sync import (
+    sync_key_access_group_membership,
+    sync_key_regeneration_access_group_membership,
+)
+
+
+def _routed_prisma_client():
+    writer_inner = MagicMock(name="writer_prisma")
+    reader_inner = MagicMock(name="reader_prisma")
+    writer_inner.query_raw = AsyncMock(return_value=[])
+    reader_inner.query_raw = AsyncMock(return_value=[])
+    writer = PrismaWrapper(original_prisma=writer_inner, iam_token_db_auth=False)
+    reader = PrismaWrapper(original_prisma=reader_inner, iam_token_db_auth=False)
+    routing = RoutingPrismaWrapper(writer=writer, reader=reader)
+    return SimpleNamespace(db=routing), writer_inner, reader_inner
+
+
+@pytest.mark.asyncio
+async def test_regeneration_repoint_update_runs_on_the_writer():
+    prisma_client, writer_inner, reader_inner = _routed_prisma_client()
+
+    await sync_key_regeneration_access_group_membership(
+        prisma_client=prisma_client,
+        previous_key_token="old-token",
+        new_key_token="new-token",
+        data=None,
+        existing_key_row=MagicMock(),
+    )
+
+    writer_inner.query_raw.assert_awaited_once()
+    assert writer_inner.query_raw.await_args.args[0].startswith('UPDATE "LiteLLM_AccessGroupTable"')
+    reader_inner.query_raw.assert_not_awaited()
+
+
+@pytest.mark.asyncio
+async def test_membership_attach_and_detach_updates_run_on_the_writer():
+    prisma_client, writer_inner, reader_inner = _routed_prisma_client()
+
+    await sync_key_access_group_membership(
+        prisma_client=prisma_client,
+        key_token="token",
+        previous_access_group_ids=["ag-old"],
+        updated_access_group_ids=["ag-new"],
+    )
+
+    assert writer_inner.query_raw.await_count == 2
+    assert all(
+        call.args[0].startswith('UPDATE "LiteLLM_AccessGroupTable"') for call in writer_inner.query_raw.await_args_list
+    )
+    reader_inner.query_raw.assert_not_awaited()
diff --git a/tests/test_litellm/proxy/proxy_server/test_routes_config.py b/tests/test_litellm/proxy/proxy_server/test_routes_config.py
index ad3c470acf3..dcb63b8ca82 100644
--- a/tests/test_litellm/proxy/proxy_server/test_routes_config.py
+++ b/tests/test_litellm/proxy/proxy_server/test_routes_config.py
@@ -60,6 +60,141 @@ def test_config_update_happy_admin(client, auth_as, mock_prisma, monkeypatch):
     assert normalize(response.json()) == {"message": "Config updated successfully"}
 
 
+def test_config_update_persists_optional_pre_call_checks(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+    fake_proxy_config = MagicMock()
+    fake_proxy_config.add_deployment = AsyncMock()
+    monkeypatch.setattr(ps, "proxy_config", fake_proxy_config)
+
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"optional_pre_call_checks": ["prompt_caching"]}},
+        )
+
+    assert response.status_code == 200
+    persisted = json.loads(table.upsert.call_args.kwargs["data"]["create"]["param_value"])
+    assert persisted["optional_pre_call_checks"] == ["prompt_caching"]
+
+
+def test_config_update_persists_model_group_affinity_config(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+    fake_proxy_config = MagicMock()
+    fake_proxy_config.add_deployment = AsyncMock()
+    monkeypatch.setattr(ps, "proxy_config", fake_proxy_config)
+
+    model_group_affinity_config = {"gpt-4": ["session_affinity"]}
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"model_group_affinity_config": model_group_affinity_config}},
+        )
+
+    assert response.status_code == 200
+    persisted = json.loads(table.upsert.call_args.kwargs["data"]["create"]["param_value"])
+    assert persisted["model_group_affinity_config"] == model_group_affinity_config
+
+
+def test_config_update_persists_disable_cooldowns(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+    fake_proxy_config = MagicMock()
+    fake_proxy_config.add_deployment = AsyncMock()
+    monkeypatch.setattr(ps, "proxy_config", fake_proxy_config)
+
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"disable_cooldowns": True}},
+        )
+
+    assert response.status_code == 200
+    persisted = json.loads(table.upsert.call_args.kwargs["data"]["create"]["param_value"])
+    assert persisted["disable_cooldowns"] is True
+
+
+def test_config_update_rejects_assistants_config(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"assistants_config": {"enabled": True}}},
+        )
+
+    assert response.status_code == 400
+    assert "assistants_config" in response.json()["error"]["message"]
+    table.upsert.assert_not_called()
+
+
+def test_config_update_rejects_router_general_settings(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"router_general_settings": {"async_only_mode": True}}},
+        )
+
+    assert response.status_code == 400
+    assert "router_general_settings" in response.json()["error"]["message"]
+    table.upsert.assert_not_called()
+
+
+def test_config_update_rejects_unknown_router_setting(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    table = _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+
+    with auth_as(LitellmUserRoles.PROXY_ADMIN):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"optional_precall_checks": ["prompt_caching"]}},
+        )
+
+    assert response.status_code == 400
+    assert "optional_precall_checks" in response.json()["error"]["message"]
+    table.upsert.assert_not_called()
+
+
+def test_config_update_unknown_router_setting_non_admin_forbidden(client, auth_as, mock_prisma, monkeypatch):
+    from litellm.proxy import proxy_server as ps
+    from litellm.proxy._types import LitellmUserRoles
+
+    _install_litellm_config(mock_prisma)
+    monkeypatch.setattr(ps, "prisma_client", mock_prisma)
+
+    with auth_as(LitellmUserRoles.INTERNAL_USER):
+        response = client.post(
+            "/config/update",
+            json={"router_settings": {"optional_precall_checks": ["prompt_caching"]}},
+        )
+
+    assert response.status_code == 403
+    assert "admin" in response.json()["error"]["message"].lower()
+
+
 def test_config_update_non_admin_forbidden(client, auth_as, mock_prisma, monkeypatch):
     """POST /config/update by a non-admin caller is rejected; the error
     surfaces as a ProxyException with the admin-only message."""
diff --git a/tests/test_litellm/proxy/proxy_server/test_routes_models.py b/tests/test_litellm/proxy/proxy_server/test_routes_models.py
index 2b126b1ea95..bc6106a06f8 100644
--- a/tests/test_litellm/proxy/proxy_server/test_routes_models.py
+++ b/tests/test_litellm/proxy/proxy_server/test_routes_models.py
@@ -45,6 +45,7 @@ def patched_models(monkeypatch):
     deployment = MagicMock()
     deployment.litellm_params.model = "gpt-4"
     router.get_deployment_by_model_group_name = MagicMock(return_value=deployment)
+    router.get_configured_display_name = MagicMock(return_value=None)
 
     monkeypatch.setattr(proxy_server, "llm_router", router)
     monkeypatch.setattr(proxy_server, "prisma_client", MagicMock())
@@ -187,6 +188,83 @@ def test_anthropic_format_carries_router_configured_token_limits(client, auth_as
     assert (claude["max_input_tokens"], claude["max_tokens"]) == (500000, 4096)
 
 
+@pytest.mark.parametrize("path", ["/v1/models", "/models"])
+def test_anthropic_format_uses_configured_display_name(client, auth_as, patched_models, path):
+    """A deployment's ``model_info.display_name`` becomes the Anthropic-native
+    ``display_name`` so Claude Code's picker shows a clean name while the id keeps
+    routing; models without one keep the id fallback, and the OpenAI-shaped
+    listing carries no display_name either way."""
+
+    def _configured(model_name):
+        return "Kimi K3" if model_name == "gpt-4" else None
+
+    patched_models.get_configured_display_name = MagicMock(side_effect=_configured)
+
+    with auth_as():
+        anthropic_response = client.get(path, headers={"anthropic-version": "2023-06-01"})
+        openai_response = client.get(path)
+
+    assert anthropic_response.status_code == 200
+    gpt_4, claude = anthropic_response.json()["data"]
+    assert (gpt_4["id"], gpt_4["display_name"]) == ("gpt-4", "Kimi K3")
+    assert (claude["id"], claude["display_name"]) == ("claude-sonnet", "claude-sonnet")
+
+    assert openai_response.status_code == 200
+    openai_models = openai_response.json()["data"]
+    assert [m["id"] for m in openai_models] == ["gpt-4", "claude-sonnet"]
+    assert all("display_name" not in m for m in openai_models)
+
+
+@pytest.mark.parametrize("params", [{}, {"scope": "expand"}])
+def test_anthropic_display_name_resolved_via_internal_team_key(
+    client, auth_as, patched_models, monkeypatch, params
+):
+    """For a team-scoped row the configured display name must be looked up by the
+    internal routing key while the entry itself is keyed by the public name, so
+    the clean name lands on the id the client actually sees."""
+    from litellm.proxy import utils as proxy_utils
+    from litellm.proxy.auth import model_checks
+
+    internal_name = "model_name_team-1_c0ffee"
+
+    patched_models.get_model_list = MagicMock(
+        return_value=[
+            {
+                "model_name": internal_name,
+                "model_info": {
+                    "team_id": "team-1",
+                    "team_public_model_name": "gpt-4-team",
+                },
+            }
+        ]
+    )
+    patched_models.get_model_names = MagicMock(return_value=[internal_name])
+    patched_models.get_configured_display_name = MagicMock(
+        side_effect=lambda model_name: "Team GPT" if model_name == internal_name else None
+    )
+
+    async def _fake_get_available_models_for_user(**kwargs):
+        return [internal_name]
+
+    monkeypatch.setattr(
+        proxy_utils,
+        "get_available_models_for_user",
+        _fake_get_available_models_for_user,
+    )
+    monkeypatch.setattr(
+        model_checks, "get_complete_model_list", lambda **kwargs: [internal_name]
+    )
+
+    with auth_as():
+        response = client.get(
+            "/v1/models", params=params, headers={"anthropic-version": "2023-06-01"}
+        )
+
+    assert response.status_code == 200
+    (entry,) = response.json()["data"]
+    assert (entry["id"], entry["display_name"]) == ("gpt-4-team", "Team GPT")
+
+
 @pytest.mark.parametrize("path", ["/v1/models", "/models"])
 def test_get_models_invalid_scope_returns_400(client, auth_as, patched_models, path):
     """Pins: ``GET /v1/models``, ``GET /models`` (error path: invalid scope)."""
diff --git a/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py b/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py
index aa35fd64f18..0fb9b1a6d88 100644
--- a/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py
+++ b/tests/test_litellm/proxy/proxy_server/test_team_model_name_translation.py
@@ -19,7 +19,10 @@ from litellm.proxy._types import (
     LitellmUserRoles,
     UserAPIKeyAuth,
 )
-from litellm.proxy.common_utils.model_listing_utils import TeamModelNameTranslator
+from litellm.proxy.common_utils.model_listing_utils import (
+    TeamModelNameTranslator,
+    configured_display_names,
+)
 from litellm.proxy.proxy_server import (
     _get_proxy_model_info,
     _translate_model_name_for_response,
@@ -1391,6 +1394,27 @@ def test_resolve_public_name_respects_legacy_flag():
     )
 
 
+def test_configured_display_names_keyed_by_response_id():
+    """The map is keyed by the public response id while the router lookup uses
+    the internal routing key, and entries without a configured name are omitted."""
+    router = MagicMock()
+    router.get_configured_display_name = MagicMock(
+        side_effect=lambda model_name: "Team Sonnet" if model_name == "model_name_team-abc-123_4a6b8" else None
+    )
+
+    assert configured_display_names(
+        entries=[
+            ("team-claude-sonnet", "model_name_team-abc-123_4a6b8"),
+            ("gpt-4o", "gpt-4o"),
+        ],
+        llm_router=router,
+    ) == {"team-claude-sonnet": "Team Sonnet"}
+
+
+def test_configured_display_names_empty_without_router():
+    assert configured_display_names(entries=[("gpt-4o", "gpt-4o")], llm_router=None) == {}
+
+
 @pytest.mark.asyncio
 async def test_retrieve_model_by_public_name_returns_200(monkeypatch):
     """Regression: `GET /v1/models/{public_name}` must NOT 404. The listing
diff --git a/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py b/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py
index 31430da71e8..8006f64ba41 100644
--- a/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py
+++ b/tests/test_litellm/proxy/public_endpoints/test_public_endpoints.py
@@ -1,3 +1,4 @@
+import re
 from datetime import datetime, timezone
 from unittest.mock import AsyncMock, MagicMock, patch
 
@@ -756,6 +757,45 @@ def test_public_agent_hub_returns_empty_when_no_public_groups():
     assert response.json() == []
 
 
+# ---------------------------------------------------------------------------
+# /public/agents/fields
+# ---------------------------------------------------------------------------
+
+
+def test_bedrock_agentcore_runtime_arn_validation_pattern_accepts_full_resource_path():
+    """Regression for LIT-6737: the AgentCore agent_runtime_arn field's
+    validation_pattern must accept a complete runtime ARN whose resource part
+    is itself multi-segment (``runtime/<runtime-id>``), and reject the exact
+    truncated shape a naive split("/")-by-position parse used to produce (the
+    ARN cut off right after the ``runtime`` resource type).
+    """
+    app_instance = FastAPI()
+    app_instance.include_router(router)
+    test_client = TestClient(app_instance)
+
+    response = test_client.get("/public/agents/fields")
+    assert response.status_code == 200
+    agents = response.json()
+
+    bedrock_agentcore = next((a for a in agents if a["agent_type"] == "bedrock_agentcore"), None)
+    assert bedrock_agentcore is not None, "bedrock_agentcore agent type not found"
+    assert bedrock_agentcore["model_template"] == "bedrock/agentcore/{agent_runtime_arn}"
+
+    fields_by_key = {f["key"]: f for f in bedrock_agentcore["credential_fields"]}
+    arn_field = fields_by_key["agent_runtime_arn"]
+    assert arn_field["required"] is True
+    assert arn_field["include_in_litellm_params"] is False
+
+    pattern = arn_field.get("validation_pattern")
+    assert pattern, "agent_runtime_arn must ship a validation_pattern so the UI can reject a truncated ARN"
+
+    full_arn = "arn:aws:bedrock-agentcore:eu-central-1:123456789012:runtime/hosted_agent_4vm3i-BaTdfOELAs"
+    truncated_arn = "arn:aws:bedrock-agentcore:eu-central-1:123456789012:runtime"
+
+    assert re.match(pattern, full_arn), "the validator must accept a complete runtime ARN"
+    assert not re.match(pattern, truncated_arn), "the validator must reject the truncated ARN"
+
+
 # ---------------------------------------------------------------------------
 # /public/endpoints
 # ---------------------------------------------------------------------------
diff --git a/tests/test_litellm/proxy/rag_endpoints/test_rag_endpoints.py b/tests/test_litellm/proxy/rag_endpoints/test_rag_endpoints.py
index abbf6892a98..0085b6ebd36 100644
--- a/tests/test_litellm/proxy/rag_endpoints/test_rag_endpoints.py
+++ b/tests/test_litellm/proxy/rag_endpoints/test_rag_endpoints.py
@@ -324,6 +324,127 @@ def test_rag_query_stream_returns_event_stream(client_internal_user):
     assert "data: [DONE]" in response.text
 
 
+def test_rag_query_merges_managed_store_params(client_internal_user):
+    """
+    Regression: /v1/rag/query must consult the managed vector store registry
+    (like the direct /v1/vector_stores/{id}/search endpoint does) so that
+    provider, region, embedding model, etc. don't have to be repeated in
+    retrieval_config. Pre-fix the registry was never read, so managed S3
+    Vectors stores failed with "aws_region_name is required".
+    """
+    import litellm
+    from litellm.types.utils import ModelResponse
+
+    mock_vector_store = {
+        "vector_store_id": "s3-store",
+        "custom_llm_provider": "s3_vectors",
+        "litellm_params": {
+            "aws_region_name": "eu-west-1",
+            "embedding_model": "my-embed",
+            "vector_bucket_name": "bkt",
+        },
+    }
+    mock_registry = MagicMock()
+    mock_registry.get_litellm_managed_vector_store_from_registry.return_value = mock_vector_store
+
+    mock_response = ModelResponse(
+        id="chatcmpl-test",
+        choices=[{"index": 0, "message": {"role": "assistant", "content": "hi"}, "finish_reason": "stop"}],
+        model="gpt-4o-mini",
+    )
+
+    with patch(  # test-quality-ok: aquery is the endpoint's downstream boundary; the forwarded config is what the test asserts
+        "litellm.proxy.rag_endpoints.endpoints.litellm.aquery",
+        new_callable=AsyncMock,
+        return_value=mock_response,
+    ) as mock_aquery, patch.object(litellm, "vector_store_registry", mock_registry), patch(  # test-quality-ok: seeds the managed-store registry the merge under test reads and grants access so real store resolution runs
+        "litellm.proxy.vector_store_endpoints.utils.can_user_access_vector_store",
+        new=AsyncMock(return_value=True),
+    ):
+        response = client_internal_user.post(
+            "/v1/rag/query",
+            json={
+                "model": "gpt-4o-mini",
+                "messages": [{"role": "user", "content": "hello"}],
+                "retrieval_config": {"vector_store_id": "s3-store"},
+            },
+        )
+
+    assert response.status_code == 200, response.json()
+    mock_aquery.assert_awaited_once()
+    forwarded_config = mock_aquery.await_args.kwargs["retrieval_config"]
+    assert forwarded_config["vector_store_id"] == "s3-store"
+    assert forwarded_config["custom_llm_provider"] == "s3_vectors"
+    assert forwarded_config["aws_region_name"] == "eu-west-1"
+    assert forwarded_config["embedding_model"] == "my-embed"
+    assert forwarded_config["vector_bucket_name"] == "bkt"
+
+
+def test_rag_query_store_params_win_over_user_retrieval_config(client_internal_user):
+    """Registry values must win over user-supplied retrieval_config keys so callers cannot override store credentials."""
+    import litellm
+    from litellm.types.utils import ModelResponse
+
+    mock_vector_store = {
+        "vector_store_id": "s3-store",
+        "custom_llm_provider": "s3_vectors",
+        "litellm_params": {"aws_region_name": "eu-west-1"},
+    }
+    mock_registry = MagicMock()
+    mock_registry.get_litellm_managed_vector_store_from_registry.return_value = mock_vector_store
+
+    mock_response = ModelResponse(
+        id="chatcmpl-test",
+        choices=[{"index": 0, "message": {"role": "assistant", "content": "hi"}, "finish_reason": "stop"}],
+        model="gpt-4o-mini",
+    )
+
+    with patch(  # test-quality-ok: aquery is the endpoint's downstream boundary; the forwarded config is what the test asserts
+        "litellm.proxy.rag_endpoints.endpoints.litellm.aquery",
+        new_callable=AsyncMock,
+        return_value=mock_response,
+    ) as mock_aquery, patch.object(litellm, "vector_store_registry", mock_registry), patch(  # test-quality-ok: seeds the managed-store registry the merge under test reads and grants access so real store resolution runs
+        "litellm.proxy.vector_store_endpoints.utils.can_user_access_vector_store",
+        new=AsyncMock(return_value=True),
+    ):
+        response = client_internal_user.post(
+            "/v1/rag/query",
+            json={
+                "model": "gpt-4o-mini",
+                "messages": [{"role": "user", "content": "hello"}],
+                "retrieval_config": {"vector_store_id": "s3-store", "aws_region_name": "us-east-1"},
+            },
+        )
+
+    assert response.status_code == 200, response.json()
+    forwarded_config = mock_aquery.await_args.kwargs["retrieval_config"]
+    assert forwarded_config["aws_region_name"] == "eu-west-1"
+
+
+@pytest.mark.parametrize(
+    "blocked_key",
+    ["embedding_model", "litellm_embedding_model", "litellm_embedding_config", "litellm_credential_name"],
+)
+def test_rag_query_rejects_caller_embedding_selection_params(client_internal_user, blocked_key):
+    """
+    Regression: a caller must not pick the embedding model or credential used at
+    search time. Those resolve through the Router with the proxy's credentials,
+    bypassing the key's model permissions, so they may only come from the
+    managed store's server-side registration.
+    """
+    response = client_internal_user.post(
+        "/v1/rag/query",
+        json={
+            "model": "gpt-4o-mini",
+            "messages": [{"role": "user", "content": "hello"}],
+            "retrieval_config": {"vector_store_id": "s3-store", blocked_key: "attacker-choice"},
+        },
+    )
+
+    assert response.status_code == 400, response.json()
+    assert blocked_key in str(response.json())
+
+
 EICAR = r"X5O!P%@AP[4\PZX54(P^)7CC)7}$EICAR-STANDARD-ANTIVIRUS-TEST-FILE!$H+H*"
 INGEST_REQUEST = '{"ingest_options":{"vector_store":{"custom_llm_provider":"openai"}}}'
 
diff --git a/tests/test_litellm/proxy/test_redis_auth_cache_flag.py b/tests/test_litellm/proxy/test_redis_auth_cache_flag.py
index 772b08bc9d0..573bfc40c96 100644
--- a/tests/test_litellm/proxy/test_redis_auth_cache_flag.py
+++ b/tests/test_litellm/proxy/test_redis_auth_cache_flag.py
@@ -5,7 +5,7 @@ Verifies that _init_cache attaches Redis to user_api_key_cache only when
 the flag is explicitly set to True, and leaves it in-memory-only otherwise.
 """
 
-from contextlib import contextmanager
+from contextlib import ExitStack, contextmanager
 import json
 from unittest.mock import MagicMock, patch
 
@@ -167,3 +167,25 @@ class TestRedisAuthCacheFlag:
                     f"cli_sso_session_cache must always get Redis "
                     f"(enable_redis_auth_cache={flag_value!r})"
                 )
+
+    def test_flag_absent_still_shares_the_model_budget_counters_over_redis(self):
+        """
+        Per-model budget counters are spend counters: the limiter must be able to
+        push and read them through Redis without the auth-cache opt-in, or every
+        worker enforces and reports its own share of a key's spend
+        """
+        fake_redis = _FakeRedisCache()
+        limiter_cache = ps.model_max_budget_limiter.dual_cache
+        touched_caches = (
+            limiter_cache,
+            ps.spend_counter_cache,
+            ps.cli_sso_session_cache,
+            ps.user_api_key_cache,
+            ps.litellm_config_cache,
+        )
+        with ExitStack() as detached:
+            for cache in touched_caches:
+                detached.enter_context(patch.object(cache, "redis_cache", None))
+            ps._attach_redis_usage_cache(fake_redis, enable_redis_auth_cache=False)
+            assert limiter_cache.redis_cache is fake_redis
+            assert ps.user_api_key_cache.redis_cache is None
diff --git a/tests/test_litellm/proxy/vector_store_endpoints/test_vector_store_endpoints.py b/tests/test_litellm/proxy/vector_store_endpoints/test_vector_store_endpoints.py
index eae6f90863a..45a0221c8a6 100644
--- a/tests/test_litellm/proxy/vector_store_endpoints/test_vector_store_endpoints.py
+++ b/tests/test_litellm/proxy/vector_store_endpoints/test_vector_store_endpoints.py
@@ -3158,3 +3158,35 @@ class TestAzureAIAnalyzeNamedIndexClassification:
             user_api_key_dict=self._team_member("analyze", ["read"]),
         )
         assert result is True
+
+
+@pytest.mark.parametrize(
+    "blocked_key",
+    ["embedding_model", "litellm_embedding_model", "litellm_embedding_config", "litellm_credential_name"],
+)
+def test_vector_store_search_rejects_caller_embedding_selection_params(blocked_key):
+    """
+    Regression: the search request body must not pick the embedding model or
+    credential used to embed the query. Those resolve through the Router with
+    the proxy's credentials, bypassing the key's model permissions, so they may
+    only come from the managed store's server-side registration.
+    """
+    from fastapi.testclient import TestClient
+
+    from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
+    from litellm.proxy.proxy_server import app
+
+    mock_auth = UserAPIKeyAuth(user_id="test_internal_user", user_role=LitellmUserRoles.INTERNAL_USER.value)
+    original_overrides = app.dependency_overrides.copy()
+    app.dependency_overrides[user_api_key_auth] = lambda: mock_auth
+    try:
+        client = TestClient(app)
+        response = client.post(
+            "/v1/vector_stores/s3-store/search",
+            json={"query": "hello", blocked_key: "attacker-choice"},
+        )
+    finally:
+        app.dependency_overrides = original_overrides
+
+    assert response.status_code == 400, response.json()
+    assert blocked_key in str(response.json())
diff --git a/tests/test_litellm/rag/test_main.py b/tests/test_litellm/rag/test_main.py
index 2d1b460513f..51d03544910 100644
--- a/tests/test_litellm/rag/test_main.py
+++ b/tests/test_litellm/rag/test_main.py
@@ -259,6 +259,135 @@ async def test_aquery_streaming_bills_sub_call_costs_into_final_event():
     assert standard_logging_object["response_cost"] >= 0.003
 
 
+@pytest.mark.asyncio
+async def test_aquery_forwards_provider_retrieval_config_and_router_to_search():
+    """
+    Regression: provider-specific retrieval_config keys (aws_region_name,
+    embedding_model, vector_bucket_name, ...) and the router must be forwarded
+    to the vector store search call. Pre-fix they were silently dropped, so
+    /v1/rag/query failed with provider config errors (e.g. S3 Vectors
+    "aws_region_name is required") even when the caller supplied them.
+    """
+    from unittest.mock import AsyncMock
+
+    from litellm.types.vector_stores import VectorStoreSearchResponse
+
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "gpt-4o-mini",
+                "litellm_params": {"model": "openai/gpt-4o-mini", "api_key": "test-key"},
+            }
+        ]
+    )
+
+    fake_search = AsyncMock(
+        return_value=VectorStoreSearchResponse(
+            object="vector_store.search_results.page", search_query="q", data=[]
+        )
+    )
+    with patch("litellm.vector_stores.asearch", new=fake_search):  # test-quality-ok: asearch is the boundary the forwarding contract under test targets
+        response = await litellm.aquery(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "hello"}],
+            retrieval_config={
+                "vector_store_id": "bkt:idx",
+                "custom_llm_provider": "s3_vectors",
+                "top_k": 5,
+                "aws_region_name": "eu-west-1",
+                "embedding_model": "my-embed",
+                "vector_bucket_name": "bkt",
+            },
+            router=router,
+            mock_response="hi",
+        )
+
+    assert isinstance(response, ModelResponse)
+    fake_search.assert_awaited_once()
+    search_kwargs = fake_search.await_args.kwargs
+    assert search_kwargs["vector_store_id"] == "bkt:idx"
+    assert search_kwargs["custom_llm_provider"] == "s3_vectors"
+    assert search_kwargs["max_num_results"] == 5
+    assert search_kwargs["router"] is router
+    # provider-specific extras forwarded
+    assert search_kwargs["aws_region_name"] == "eu-west-1"
+    assert search_kwargs["embedding_model"] == "my-embed"
+    assert search_kwargs["vector_bucket_name"] == "bkt"
+    # consumed keys are not duplicated into the spread
+    assert "top_k" not in search_kwargs
+
+
+@pytest.mark.asyncio
+async def test_aquery_minimal_retrieval_config_forwards_no_extras():
+    """
+    A minimal retrieval_config must not leak consumed keys (or invent extras)
+    into the vector store search call.
+    """
+    from unittest.mock import AsyncMock
+
+    from litellm.types.vector_stores import VectorStoreSearchResponse
+
+    fake_search = AsyncMock(
+        return_value=VectorStoreSearchResponse(
+            object="vector_store.search_results.page", search_query="q", data=[]
+        )
+    )
+    with patch("litellm.vector_stores.asearch", new=fake_search):  # test-quality-ok: asearch is the boundary the forwarding contract under test targets
+        await litellm.aquery(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "hello"}],
+            retrieval_config={"vector_store_id": "vs_test_123", "custom_llm_provider": "openai"},
+            mock_response="hi",
+        )
+
+    fake_search.assert_awaited_once()
+    search_kwargs = fake_search.await_args.kwargs
+    assert search_kwargs["vector_store_id"] == "vs_test_123"
+    assert search_kwargs["custom_llm_provider"] == "openai"
+    assert search_kwargs["router"] is None
+    leaked = {"top_k", "filters", "retrieval_filter", "aws_region_name", "embedding_model", "vector_bucket_name"}
+    assert not (leaked & set(search_kwargs.keys()))
+
+
+@pytest.mark.asyncio
+async def test_aquery_does_not_forward_connection_override_keys_to_search():
+    """
+    Only allowlisted retrieval_config keys may reach the vector store search
+    call. Caller-controlled connection overrides (api_base, api_key, arbitrary
+    extras) must be dropped, otherwise a caller could redirect store
+    credentials to an attacker-chosen host.
+    """
+    from unittest.mock import AsyncMock
+
+    from litellm.types.vector_stores import VectorStoreSearchResponse
+
+    fake_search = AsyncMock(
+        return_value=VectorStoreSearchResponse(
+            object="vector_store.search_results.page", search_query="q", data=[]
+        )
+    )
+    with patch("litellm.vector_stores.asearch", new=fake_search):  # test-quality-ok: asearch is the boundary the forwarding contract under test targets
+        await litellm.aquery(
+            model="gpt-4o-mini",
+            messages=[{"role": "user", "content": "hello"}],
+            retrieval_config={
+                "vector_store_id": "bkt:idx",
+                "custom_llm_provider": "s3_vectors",
+                "aws_region_name": "eu-west-1",
+                "api_base": "https://attacker.example.com",
+                "api_key": "attacker-key",
+                "arbitrary_extra": "nope",
+            },
+            mock_response="hi",
+        )
+
+    fake_search.assert_awaited_once()
+    search_kwargs = fake_search.await_args.kwargs
+    assert search_kwargs["aws_region_name"] == "eu-west-1"
+    blocked = {"api_base", "api_key", "arbitrary_extra"}
+    assert not (blocked & set(search_kwargs.keys()))
+
+
 def test_rag_call_types_are_registered():
     """
     query/aquery/ingest/aingest are @client-decorated entry points, so their
diff --git a/tests/test_litellm/rerank_api/test_main.py b/tests/test_litellm/rerank_api/test_main.py
index 587be59c550..2b6cfeda2c2 100644
--- a/tests/test_litellm/rerank_api/test_main.py
+++ b/tests/test_litellm/rerank_api/test_main.py
@@ -111,6 +111,99 @@ def test_together_rerank_honors_api_base(respx_mock: respx.MockRouter):
     assert mock_route.calls[0].request.headers["authorization"] == "Bearer fake-together-key"
 
 
+DASHSCOPE_404_BODY = {
+    "error": {
+        "message": "The model `does-not-exist` does not exist or you do not have access to it.",
+        "type": "invalid_request_error",
+        "param": None,
+        "code": "model_not_found",
+    },
+    "request_id": "mock-request-id",
+}
+
+
+def test_rerank_error_names_provider_and_keeps_body(respx_mock: respx.MockRouter, monkeypatch):
+    """Regression for the rerank error path mapping with the unresolved provider param:
+    a provider 404 surfaced as 'None - ' instead of naming the provider and its error body."""
+    monkeypatch.delenv("DASHSCOPE_API_BASE", raising=False)
+    monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False)
+
+    mock_route = respx_mock.post("https://dashscope.example/v1/reranks")
+    mock_route.return_value = httpx.Response(404, json=DASHSCOPE_404_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as exc_info:
+        litellm.rerank(
+            model="dashscope/does-not-exist",
+            query=MARKER_QUERY,
+            documents=[MARKER_DOC],
+            api_key="fake-dashscope-key",
+            api_base="https://dashscope.example/v1",
+        )
+
+    assert mock_route.called
+    assert "DashscopeException" in str(exc_info.value)
+    assert "does not exist or you do not have access to it" in str(exc_info.value)
+    assert "None - " not in str(exc_info.value)
+
+
+@pytest.mark.asyncio
+async def test_arerank_error_is_mapped_to_litellm_exception(respx_mock: respx.MockRouter, monkeypatch):
+    """Regression for arerank's bare re-raise: provider errors escaped as raw
+    provider exception classes instead of the mapped litellm exception contract."""
+    monkeypatch.delenv("DASHSCOPE_API_BASE", raising=False)
+    monkeypatch.delenv("DASHSCOPE_API_BASE_RERANK", raising=False)
+    monkeypatch.setenv("DISABLE_AIOHTTP_TRANSPORT", "True")
+
+    mock_route = respx_mock.post("https://dashscope.example/v1/reranks")
+    mock_route.return_value = httpx.Response(404, json=DASHSCOPE_404_BODY)
+
+    with pytest.raises(litellm.NotFoundError) as exc_info:
+        await litellm.arerank(
+            model="dashscope/does-not-exist",
+            query=MARKER_QUERY,
+            documents=[MARKER_DOC],
+            api_key="fake-dashscope-key",
+            api_base="https://dashscope.example/v1",
+        )
+
+    assert mock_route.called
+    assert "DashscopeException" in str(exc_info.value)
+    assert "does not exist or you do not have access to it" in str(exc_info.value)
+    assert "None - " not in str(exc_info.value)
+
+
+@pytest.mark.asyncio
+async def test_arerank_declared_authenticating_provider_skips_resolution(monkeypatch):
+    """Regression for the event-loop hazard in arerank's provider pre-resolution:
+    get_llm_provider runs the blocking OAuth device flow for github_copilot/chatgpt,
+    so arerank must adopt the declared provider instead of resolving it, while the
+    except path still maps with that declared provider."""
+    from litellm.llms.base_llm.chat.transformation import BaseLLMException
+
+    resolution_calls = []
+
+    def record_resolution(*args, **kwargs):
+        resolution_calls.append((args, kwargs))
+        return "gpt-4o", "github_copilot", None, None
+
+    def rerank_raises_provider_error(*args, **kwargs):
+        raise BaseLLMException(status_code=401, message='{"error":"bad key"}')
+
+    monkeypatch.setattr(litellm, "get_llm_provider", record_resolution)
+    monkeypatch.setattr("litellm.rerank_api.main.rerank", rerank_raises_provider_error)
+
+    with pytest.raises(litellm.AuthenticationError) as exc_info:
+        await litellm.arerank(
+            model="github_copilot/gpt-4o",
+            query=MARKER_QUERY,
+            documents=[MARKER_DOC],
+        )
+
+    assert resolution_calls == []
+    assert "Github_copilotException" in str(exc_info.value)
+    assert "None - " not in str(exc_info.value)
+
+
 @pytest.mark.asyncio
 async def test_together_rerank_async_honors_env_api_base(respx_mock: respx.MockRouter, monkeypatch):
     """Regression: TOGETHER_AI_API_BASE was honored by chat but ignored by rerank."""
diff --git a/tests/test_litellm/responses/test_streaming_iterator.py b/tests/test_litellm/responses/test_streaming_iterator.py
index 677faf7f655..9edcaaef034 100644
--- a/tests/test_litellm/responses/test_streaming_iterator.py
+++ b/tests/test_litellm/responses/test_streaming_iterator.py
@@ -326,3 +326,55 @@ def test_run_post_success_hooks_does_not_report_generation_time_as_overhead():
 
     assert iterator.completed_response._hidden_params["_response_ms"] == 10000.0
     assert "litellm_overhead_time_ms" not in iterator.completed_response._hidden_params
+
+
+def _responses_api_response_with_usage() -> ResponsesAPIResponse:
+    from litellm.types.llms.openai import ResponseAPIUsage
+
+    return ResponsesAPIResponse(
+        id="resp_lit6427",
+        created_at=int(datetime(2025, 1, 1).timestamp()),
+        status="completed",
+        model="mantle-claude",
+        object="response",
+        output=[],
+        usage=ResponseAPIUsage(input_tokens=20, output_tokens=60, total_tokens=80),
+    )
+
+
+def test_stamp_responses_usage_cost_stamps_computed_cost():
+    from litellm.responses.streaming_iterator import _stamp_responses_usage_cost
+
+    response = _responses_api_response_with_usage()
+    logging_obj = Mock(spec=LiteLLMLoggingObj)
+    logging_obj._response_cost_calculator.return_value = 0.000704
+
+    _stamp_responses_usage_cost(response, logging_obj)
+
+    assert getattr(response.usage, "cost", None) == pytest.approx(0.000704)
+    logging_obj._response_cost_calculator.assert_called_once_with(result=response)
+
+
+def test_stamp_responses_usage_cost_keeps_provider_reported_cost():
+    from litellm.responses.streaming_iterator import _stamp_responses_usage_cost
+
+    response = _responses_api_response_with_usage()
+    setattr(response.usage, "cost", 0.5)
+    logging_obj = Mock(spec=LiteLLMLoggingObj)
+
+    _stamp_responses_usage_cost(response, logging_obj)
+
+    assert getattr(response.usage, "cost", None) == pytest.approx(0.5)
+    logging_obj._response_cost_calculator.assert_not_called()
+
+
+def test_stamp_responses_usage_cost_survives_calculator_failure():
+    from litellm.responses.streaming_iterator import _stamp_responses_usage_cost
+
+    response = _responses_api_response_with_usage()
+    logging_obj = Mock(spec=LiteLLMLoggingObj)
+    logging_obj._response_cost_calculator.side_effect = RuntimeError("cost map unavailable")
+
+    _stamp_responses_usage_cost(response, logging_obj)
+
+    assert getattr(response.usage, "cost", None) is None
diff --git a/tests/test_litellm/rust_bridge/test_verify_linux_native_wheel.py b/tests/test_litellm/rust_bridge/test_verify_linux_native_wheel.py
new file mode 100644
index 00000000000..e449d4392d8
--- /dev/null
+++ b/tests/test_litellm/rust_bridge/test_verify_linux_native_wheel.py
@@ -0,0 +1,197 @@
+from __future__ import annotations
+
+import importlib.util
+import subprocess
+import sys
+import zipfile
+from collections.abc import Callable, Mapping, Sequence
+from pathlib import Path
+from types import MappingProxyType, ModuleType
+from typing import Final, Protocol, cast
+
+import pytest
+
+
+class _CommandRunner(Protocol):
+    def __call__(
+        self,
+        command: tuple[str, ...],
+        *,
+        check: bool,
+        capture_output: bool,
+        text: bool,
+    ) -> subprocess.CompletedProcess[str]: ...
+
+
+class _VerifierModule(Protocol):
+    main: Callable[
+        [
+            Sequence[str] | None,
+            Mapping[str, str] | None,
+            Callable[[Path], ModuleType | None],
+            _CommandRunner,
+        ],
+        int,
+    ]
+
+
+_REPO_ROOT: Final = Path(__file__).resolve().parents[3]
+_MODULE_PATH: Final = _REPO_ROOT / ".github" / "scripts" / "verify_linux_native_wheel.py"
+_SPEC: Final = importlib.util.spec_from_file_location("verify_linux_native_wheel", _MODULE_PATH)
+assert _SPEC is not None and _SPEC.loader is not None
+_LOADED_VERIFIER: Final = importlib.util.module_from_spec(_SPEC)
+sys.modules[_SPEC.name] = _LOADED_VERIFIER
+_SPEC.loader.exec_module(_LOADED_VERIFIER)
+verifier: Final = cast(_VerifierModule, _LOADED_VERIFIER)
+
+_EXPECTED_TAG: Final = "cp310-abi3-linux_x86_64"
+_NATIVE_MEMBER: Final = "litellm/rust_bridge/_native.abi3.so"
+_DIST_INFO: Final = "litellm-1.100.0.dist-info"
+
+
+def _write_wheel(
+    tmp_path: Path,
+    *,
+    filename_tag: str,
+    metadata_tags: tuple[str, ...] | None = (_EXPECTED_TAG,),
+    dist_info: str = _DIST_INFO,
+    duplicate_wheel: bool = False,
+) -> Path:
+    wheel: Final = tmp_path / f"litellm-1.100.0-{filename_tag}.whl"
+    with zipfile.ZipFile(wheel, "w", compression=zipfile.ZIP_DEFLATED) as archive:
+        archive.writestr(_NATIVE_MEMBER, b"synthetic native extension")
+        archive.writestr(
+            f"{dist_info}/METADATA",
+            "Metadata-Version: 2.1\nName: litellm\nVersion: 1.100.0\n",
+        )
+        archive.writestr(
+            f"{dist_info}/RECORD",
+            f"{_NATIVE_MEMBER},,\n{dist_info}/WHEEL,,\n",
+        )
+        if metadata_tags is not None:
+            wheel_metadata: Final = (
+                "Wheel-Version: 1.0\nGenerator: regression-test\nRoot-Is-Purelib: false\n"
+                + "".join(f"Tag: {tag}\n" for tag in metadata_tags)
+            )
+            archive.writestr(f"{dist_info}/WHEEL", wheel_metadata)
+            if duplicate_wheel:
+                archive.writestr(f"{dist_info}/WHEEL", wheel_metadata)
+    return wheel
+
+
+def _fake_subprocess_run(
+    command: tuple[str, ...],
+    *,
+    check: bool,
+    capture_output: bool,
+    text: bool,
+) -> subprocess.CompletedProcess[str]:
+    assert check and capture_output and text
+    if command == ("rustc", "--version"):
+        return subprocess.CompletedProcess(command, 0, stdout="rustc 1.98.0 (regression-test)\n", stderr="")
+    if "--sections" in command:
+        return subprocess.CompletedProcess(command, 0, stdout="[ 1] .text PROGBITS\n", stderr="")
+    if "--dyn-syms" in command:
+        return subprocess.CompletedProcess(command, 0, stdout="PyInit__native\n", stderr="")
+    raise AssertionError(f"unexpected subprocess command: {command}")
+
+
+class _NativeModuleWithPanicHook(ModuleType):
+    def _panic_for_test(self) -> None:
+        return None
+
+
+def _run_verifier(
+    wheel: Path,
+    *,
+    exposes_panic: bool = False,
+) -> int:
+    native_module: Final = (
+        _NativeModuleWithPanicHook("litellm.rust_bridge._native")
+        if exposes_panic
+        else ModuleType("litellm.rust_bridge._native")
+    )
+
+    def _fake_load_native_module(_: Path) -> ModuleType:
+        return native_module
+
+    environment: Final = MappingProxyType({"GITHUB_STEP_SUMMARY": str(wheel.parent / "summary.md")})
+    return verifier.main(
+        (str(_MODULE_PATH), str(wheel)),
+        environment,
+        _fake_load_native_module,
+        _fake_subprocess_run,
+    )
+
+
+def test_accepts_expected_release_wheel_tags(tmp_path: Path) -> None:
+    wheel: Final = _write_wheel(tmp_path, filename_tag=_EXPECTED_TAG)
+
+    assert _run_verifier(wheel) == 0
+
+
+def test_rejects_cp312_version_specific_wheel(tmp_path: Path) -> None:
+    tag: Final = "cp312-cp312-linux_x86_64"
+    wheel: Final = _write_wheel(tmp_path, filename_tag=tag, metadata_tags=(tag,))
+
+    assert _run_verifier(wheel) == 1
+
+
+def test_rejects_non_linux_platform_tag(tmp_path: Path) -> None:
+    tag: Final = "cp310-abi3-win_amd64"
+    wheel: Final = _write_wheel(tmp_path, filename_tag=tag, metadata_tags=(tag,))
+
+    assert _run_verifier(wheel) == 1
+
+
+@pytest.mark.parametrize(
+    "metadata_tags",
+    (None, ("cp312-cp312-linux_x86_64",)),
+    ids=("missing", "mismatched"),
+)
+def test_rejects_missing_or_mismatched_wheel_metadata_tag(
+    tmp_path: Path,
+    metadata_tags: tuple[str, ...] | None,
+) -> None:
+    wheel: Final = _write_wheel(tmp_path, filename_tag=_EXPECTED_TAG, metadata_tags=metadata_tags)
+
+    assert _run_verifier(wheel) == 1
+
+
+def test_rejects_wheel_metadata_from_wrong_dist_info_directory(
+    tmp_path: Path,
+) -> None:
+    wheel: Final = _write_wheel(
+        tmp_path,
+        filename_tag=_EXPECTED_TAG,
+        dist_info="decoy-1.0.0.dist-info",
+    )
+
+    assert _run_verifier(wheel) == 1
+
+
+def test_rejects_duplicate_wheel_metadata_tags(tmp_path: Path) -> None:
+    wheel: Final = _write_wheel(
+        tmp_path,
+        filename_tag=_EXPECTED_TAG,
+        metadata_tags=(_EXPECTED_TAG, _EXPECTED_TAG),
+    )
+
+    assert _run_verifier(wheel) == 1
+
+
+def test_rejects_duplicate_wheel_metadata_file(tmp_path: Path) -> None:
+    with pytest.warns(UserWarning, match="Duplicate name"):
+        wheel: Final = _write_wheel(
+            tmp_path,
+            filename_tag=_EXPECTED_TAG,
+            duplicate_wheel=True,
+        )
+
+    assert _run_verifier(wheel) == 1
+
+
+def test_rejects_production_module_exposing_panic_hook(tmp_path: Path) -> None:
+    wheel: Final = _write_wheel(tmp_path, filename_tag=_EXPECTED_TAG)
+
+    assert _run_verifier(wheel, exposes_panic=True) == 1
diff --git a/tests/test_litellm/test_bedrock_usgov_pricing.py b/tests/test_litellm/test_bedrock_usgov_pricing.py
index 6b3312b5cc4..f7d95ecda01 100644
--- a/tests/test_litellm/test_bedrock_usgov_pricing.py
+++ b/tests/test_litellm/test_bedrock_usgov_pricing.py
@@ -26,9 +26,7 @@ import pytest
 
 @pytest.fixture(scope="module")
 def model_data():
-    json_path = os.path.join(
-        os.path.dirname(__file__), "../../model_prices_and_context_window.json"
-    )
+    json_path = os.path.join(os.path.dirname(__file__), "../../model_prices_and_context_window.json")
     with open(json_path) as f:
         return json.load(f)
 
@@ -51,21 +49,14 @@ def test_usgov_sonnet_4_5_pricing(model_data, model_key):
     info = model_data[model_key]
 
     assert info["input_cost_per_token"] == 3.6e-06, (
-        f"{model_key}: input_cost_per_token should be $3.60/MTok "
-        f"(got {info['input_cost_per_token']})"
+        f"{model_key}: input_cost_per_token should be $3.60/MTok (got {info['input_cost_per_token']})"
     )
-    assert (
-        info["output_cost_per_token"] == 1.8e-05
-    ), f"{model_key}: output_cost_per_token should be $18.00/MTok"
-    assert (
-        info["cache_creation_input_token_cost"] == 4.5e-06
-    ), f"{model_key}: 5m cache write should be $4.50/MTok"
-    assert (
-        info["cache_creation_input_token_cost_above_1hr"] == 7.2e-06
-    ), f"{model_key}: 1h cache write should be $7.20/MTok"
-    assert (
-        info["cache_read_input_token_cost"] == 3.6e-07
-    ), f"{model_key}: cache read should be $0.36/MTok"
+    assert info["output_cost_per_token"] == 1.8e-05, f"{model_key}: output_cost_per_token should be $18.00/MTok"
+    assert info["cache_creation_input_token_cost"] == 4.5e-06, f"{model_key}: 5m cache write should be $4.50/MTok"
+    assert info["cache_creation_input_token_cost_above_1hr"] == 7.2e-06, (
+        f"{model_key}: 1h cache write should be $7.20/MTok"
+    )
+    assert info["cache_read_input_token_cost"] == 3.6e-07, f"{model_key}: cache read should be $0.36/MTok"
 
 
 def test_usgov_carries_20_percent_premium_over_global(model_data):
@@ -84,9 +75,7 @@ def test_usgov_carries_20_percent_premium_over_global(model_data):
         "cache_read_input_token_cost",
     ):
         ratio = usgov_info[field] / global_info[field]
-        assert (
-            abs(ratio - 1.2) < 1e-9
-        ), f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
+        assert abs(ratio - 1.2) < 1e-9, f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
 
 
 # The us-gov.anthropic.* cross-region inference profile is the only us-gov
@@ -112,9 +101,7 @@ def test_usgov_cross_region_above_200k_carries_gov_premium(model_data, field, ex
     """
     info = model_data[USGOV_CROSS_REGION_KEY]
     assert field in info, f"{USGOV_CROSS_REGION_KEY}: missing field {field}"
-    assert (
-        info[field] == expected
-    ), f"{USGOV_CROSS_REGION_KEY}: {field} should be {expected} (got {info[field]})"
+    assert info[field] == expected, f"{USGOV_CROSS_REGION_KEY}: {field} should be {expected} (got {info[field]})"
 
 
 def test_usgov_cross_region_above_200k_ratio_to_global(model_data):
@@ -127,6 +114,176 @@ def test_usgov_cross_region_above_200k_ratio_to_global(model_data):
     usgov_info = model_data[USGOV_CROSS_REGION_KEY]
     for field in EXPECTED_USGOV_ABOVE_200K:
         ratio = usgov_info[field] / global_info[field]
-        assert (
-            abs(ratio - 1.2) < 1e-9
-        ), f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
+        assert abs(ratio - 1.2) < 1e-9, f"{field}: us-gov / global ratio is {ratio}, expected 1.2"
+
+
+CLAUDE_GOV_EXPECTED = {
+    "anthropic.claude-sonnet-5": {
+        "input_cost_per_token": 2.4e-06,
+        "output_cost_per_token": 1.2e-05,
+        "cache_creation_input_token_cost": 3e-06,
+        "cache_creation_input_token_cost_above_1hr": 4.8e-06,
+        "cache_read_input_token_cost": 2.4e-07,
+    },
+    "anthropic.claude-opus-4-8": {
+        "input_cost_per_token": 6e-06,
+        "output_cost_per_token": 3e-05,
+        "cache_creation_input_token_cost": 7.5e-06,
+        "cache_creation_input_token_cost_above_1hr": 1.2e-05,
+        "cache_read_input_token_cost": 6e-07,
+    },
+}
+
+
+USGOV_CLAUDE_KEY_TEMPLATES = {
+    "bedrock/us-gov-east-1/{base_key}": "bedrock",
+    "bedrock/us-gov-west-1/{base_key}": "bedrock",
+    "us-gov.{base_key}": "bedrock_converse",
+}
+
+
+@pytest.mark.parametrize("base_key", CLAUDE_GOV_EXPECTED)
+@pytest.mark.parametrize("key_template,expected_provider", USGOV_CLAUDE_KEY_TEMPLATES.items())
+def test_usgov_claude_sonnet5_opus48_pricing(model_data, key_template, expected_provider, base_key):
+    """Sonnet 5 and Opus 4.8 gov entries, both in-region keys and the us-gov.
+    geo inference profile the model cards list for GovCloud, must match the
+    rates AWS publishes on the Bedrock pricing page (1.2x global).
+    """
+    gov_key = key_template.format(base_key=base_key)
+    assert gov_key in model_data, f"Missing model entry: {gov_key}"
+    info = model_data[gov_key]
+    assert info["litellm_provider"] == expected_provider
+    for field, expected in CLAUDE_GOV_EXPECTED[base_key].items():
+        assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
+        ratio = info[field] / model_data[base_key][field]
+        assert abs(ratio - 1.2) < 1e-9, f"{gov_key}: {field} gov/global ratio is {ratio}, expected 1.2"
+
+
+CONVERSE_GOV_EXPECTED = {
+    "nvidia.nemotron-nano-3-30b": (7.2e-08, 2.88e-07),
+    "nvidia.nemotron-nano-12b-v2": (2.4e-07, 7.2e-07),
+    "nvidia.nemotron-super-3-120b": (1.8e-07, 7.8e-07),
+    "openai.gpt-oss-20b-1:0": (8.4e-08, 3.6e-07),
+    "openai.gpt-oss-120b-1:0": (1.8e-07, 7.2e-07),
+}
+
+
+@pytest.mark.parametrize("base_key", CONVERSE_GOV_EXPECTED)
+@pytest.mark.parametrize("region", ["us-gov-east-1", "us-gov-west-1"])
+def test_usgov_converse_model_pricing(model_data, region, base_key):
+    """Nemotron and gpt-oss gov entries must match the AWS Bedrock offer file,
+    which prices both GovCloud regions identically at 1.2x commercial.
+    """
+    gov_key = f"bedrock/{region}/{base_key}"
+    assert gov_key in model_data, f"Missing model entry: {gov_key}"
+    info = model_data[gov_key]
+    expected_input, expected_output = CONVERSE_GOV_EXPECTED[base_key]
+    assert info["input_cost_per_token"] == expected_input
+    assert info["output_cost_per_token"] == expected_output
+    assert info["litellm_provider"] == "bedrock"
+    base = model_data[base_key]
+    assert abs(info["input_cost_per_token"] / base["input_cost_per_token"] - 1.2) < 1e-9
+    assert abs(info["output_cost_per_token"] / base["output_cost_per_token"] - 1.2) < 1e-9
+
+
+def test_usgov_west_llama3_8b_output_price_fixed(model_data):
+    """The us-gov-west-1 llama3-8b entry carried the 70B output rate ($2.65/MTok);
+    the AWS Bedrock offer file prices output at $0.60/MTok. AWS lists the model
+    in us-gov-west-1 only, so there is no east entry to check.
+    """
+    info = model_data["bedrock/us-gov-west-1/meta.llama3-8b-instruct-v1:0"]
+    assert info["input_cost_per_token"] == 3e-07
+    assert info["output_cost_per_token"] == 6e-07
+
+
+MANTLE_GOV_TIERED_EXPECTED = {
+    "openai.gpt-5.6-luna": {
+        "input_cost_per_token": 2.64e-07,
+        "input_cost_per_token_above_272k_tokens": 5.28e-07,
+        "cache_creation_input_token_cost": 3.3e-07,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-07,
+        "cache_read_input_token_cost": 2.64e-08,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-08,
+        "output_cost_per_token": 1.584e-06,
+        "output_cost_per_token_above_272k_tokens": 2.376e-06,
+    },
+    "openai.gpt-5.6-terra": {
+        "input_cost_per_token": 2.64e-06,
+        "input_cost_per_token_above_272k_tokens": 5.28e-06,
+        "cache_creation_input_token_cost": 3.3e-06,
+        "cache_creation_input_token_cost_above_272k_tokens": 6.6e-06,
+        "cache_read_input_token_cost": 2.64e-07,
+        "cache_read_input_token_cost_above_272k_tokens": 5.28e-07,
+        "output_cost_per_token": 1.584e-05,
+        "output_cost_per_token_above_272k_tokens": 2.376e-05,
+    },
+}
+
+
+@pytest.mark.parametrize("model", MANTLE_GOV_TIERED_EXPECTED)
+def test_usgov_west_mantle_terra_luna_pricing(model_data, model):
+    """Terra and Luna carry 1.2x commercial across every tier in the
+    us-gov-west-1 offer file; the us-gov-east-1 offer file has no SKUs for them.
+    """
+    gov_key = f"bedrock_mantle/us-gov-west-1/{model}"
+    assert gov_key in model_data, f"Missing model entry: {gov_key}"
+    info = model_data[gov_key]
+    for field, expected in MANTLE_GOV_TIERED_EXPECTED[model].items():
+        assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
+    assert info["litellm_provider"] == "bedrock_mantle"
+    assert f"bedrock_mantle/us-gov-east-1/{model}" not in model_data
+
+
+@pytest.mark.parametrize("region", ["us-gov-east-1", "us-gov-west-1"])
+def test_usgov_mantle_gpt_5_4_pricing_has_no_long_context_tier(model_data, region):
+    """gpt-5.4 gov rates come from the offer file, which publishes only the
+    standard tier in GovCloud: no long-context SKUs exist there, unlike commercial.
+    """
+    gov_key = f"bedrock_mantle/{region}/openai.gpt-5.4"
+    assert gov_key in model_data, f"Missing model entry: {gov_key}"
+    info = model_data[gov_key]
+    assert info["input_cost_per_token"] == 3.3e-06
+    assert info["cache_read_input_token_cost"] == 3.3e-07
+    assert info["output_cost_per_token"] == 1.98e-05
+    assert not any(field.endswith("_above_272k_tokens") for field in info)
+
+
+def test_usgov_mantle_grok_4_3_west_only(model_data):
+    """grok-4.3 is priced in the us-gov-west-1 offer file only; the east offer
+    file carries grok-4.6 instead.
+    """
+    info = model_data["bedrock_mantle/us-gov-west-1/xai.grok-4.3"]
+    assert info["input_cost_per_token"] == 1.5e-06
+    assert info["output_cost_per_token"] == 3e-06
+    assert info["cache_read_input_token_cost"] == 2.4e-07
+    assert "bedrock_mantle/us-gov-east-1/xai.grok-4.3" not in model_data
+
+
+AZURE_GOV_EXPECTED = {
+    "azure/us-gov/gpt-5.1": {
+        "input_cost_per_token": 1.71875e-06,
+        "cache_read_input_token_cost": 1.71875e-07,
+        "output_cost_per_token": 1.375e-05,
+    },
+    "azure/us-gov/o3-mini": {
+        "input_cost_per_token": 1.513e-06,
+        "cache_read_input_token_cost": 7.57e-07,
+        "output_cost_per_token": 6.05e-06,
+    },
+    "azure/us-gov/text-embedding-3-large": {"input_cost_per_token": 1.63e-07},
+    "azure/us-gov/text-embedding-3-small": {"input_cost_per_token": 2.5e-08},
+}
+
+
+@pytest.mark.parametrize("gov_key", AZURE_GOV_EXPECTED)
+def test_azure_usgov_pricing(model_data, gov_key):
+    """Azure Government meters from the Azure retail prices API
+    (usgovvirginia/usgovarizona, serviceName 'Foundry Models'). No Government
+    retirement schedule is published, so these entries carry no deprecation_date.
+    """
+    assert gov_key in model_data, f"Missing model entry: {gov_key}"
+    info = model_data[gov_key]
+    for field, expected in AZURE_GOV_EXPECTED[gov_key].items():
+        assert info[field] == expected, f"{gov_key}: {field} should be {expected} (got {info[field]})"
+    assert info["litellm_provider"] == "azure"
+    assert "deprecation_date" not in info
diff --git a/tests/test_litellm/test_circleci_rust_toolchain.py b/tests/test_litellm/test_circleci_rust_toolchain.py
index c35ced51e16..800ca21b95d 100644
--- a/tests/test_litellm/test_circleci_rust_toolchain.py
+++ b/tests/test_litellm/test_circleci_rust_toolchain.py
@@ -17,28 +17,27 @@ Two invariants are pinned here:
      Windows job, so the check accepts either. A new job that syncs without one
      falls back to the unpinned path, which is exactly the regression a static
      check catches at PR time and a green CI run does not.
-  2. `install_rust` itself pins what it downloads: an explicit rustup version in
-     the URL, a verified SHA-256, and an exact toolchain version rather than a
-     channel name.
-
-The Windows job predates `install_rust` and provisions its toolchain inline, so
-invariant 2 is scoped to `install_rust`; invariant 1 covers both.
+  2. Both installers pin what they download: an explicit rustup version, a
+     verified SHA-256, and the exact toolchain in `rust-toolchain.toml`.
 """
 
 from __future__ import annotations
 
 import re
 from pathlib import Path
+from typing import Final
 
 import pytest
 import yaml
 
 REPO_ROOT = Path(__file__).resolve().parents[2]
 CONFIG = REPO_ROOT / ".circleci" / "config.yml"
+TOOLCHAIN: Final = REPO_ROOT / "rust-toolchain.toml"
 
 BUILDS_WORKSPACE = re.compile(r"\buv\s+(?:sync|build)\b")
 RUSTUP_ARCHIVE_URL = re.compile(r"https://static\.rust-lang\.org/rustup/archive/\d+\.\d+\.\d+/")
-EXACT_TOOLCHAIN = re.compile(r"--default-toolchain\s+\"?\d+\.\d+\.\d+\"?")
+EXACT_TOOLCHAIN = re.compile(r"--default-toolchain\s+\"?(\d+\.\d+\.\d+)\"?")
+TOOLCHAIN_CHANNEL: Final = re.compile(r'^channel = "(\d+\.\d+\.\d+)"$', re.MULTILINE)
 
 
 def _config() -> dict[str, object]:
@@ -57,6 +56,12 @@ def _step_text(step: object) -> str:
     return ""
 
 
+def _pinned_toolchain() -> str:
+    match: Final = TOOLCHAIN_CHANNEL.search(TOOLCHAIN.read_text())
+    assert match is not None, "rust-toolchain.toml must pin an exact channel"
+    return match.group(1)
+
+
 def _without_comments(text: str) -> str:
     return "\n".join(line for line in text.splitlines() if not line.lstrip().startswith("#"))
 
@@ -142,7 +147,17 @@ def test_install_rust_verifies_the_installer_checksum(install_rust_command: str)
 
 
 def test_install_rust_pins_an_exact_toolchain_version(install_rust_command: str) -> None:
-    assert EXACT_TOOLCHAIN.search(install_rust_command), (
-        "install_rust must pin an exact toolchain version (e.g. 1.97.1); a channel name like "
+    match: Final = EXACT_TOOLCHAIN.search(install_rust_command)
+    assert match is not None, (
+        "install_rust must pin an exact toolchain version (e.g. 1.98.0); a channel name like "
         "stable/beta/nightly makes the compiler drift with whatever upstream published that day"
     )
+    assert match.group(1) == _pinned_toolchain()
+
+
+def test_windows_installer_matches_the_repo_toolchain() -> None:
+    windows_steps: Final = _step_lists()["job using_litellm_on_windows"]
+    windows_command: Final = "\n".join(_step_text(step) for step in windows_steps)
+    match: Final = EXACT_TOOLCHAIN.search(windows_command)
+    assert match is not None
+    assert match.group(1) == _pinned_toolchain()
diff --git a/tests/test_litellm/test_cloudflare_workers_ai_model_metadata.py b/tests/test_litellm/test_cloudflare_workers_ai_model_metadata.py
index 9ca4515239a..e33bcfb8378 100644
--- a/tests/test_litellm/test_cloudflare_workers_ai_model_metadata.py
+++ b/tests/test_litellm/test_cloudflare_workers_ai_model_metadata.py
@@ -75,6 +75,22 @@ def test_additional_current_models_are_present():
         assert entry["output_cost_per_token"] > 0
 
 
+@pytest.mark.parametrize(
+    "key, published_price_per_audio_minute",
+    [
+        ("cloudflare/@cf/openai/whisper", 0.00045),
+        ("cloudflare/@cf/openai/whisper-large-v3-turbo", 0.00051),
+    ],
+)
+def test_whisper_transcription_pricing_is_stored_per_second(key, published_price_per_audio_minute):
+    entry = litellm.model_cost[key]
+    assert entry["litellm_provider"] == "cloudflare"
+    assert entry["mode"] == "audio_transcription"
+    assert entry["supported_endpoints"] == ["/v1/audio/transcriptions"]
+    assert entry["output_cost_per_second"] == 0.0
+    assert entry["input_cost_per_second"] == pytest.approx(published_price_per_audio_minute / 60)
+
+
 def test_root_and_backup_have_identical_cloudflare_keys():
     if not os.path.exists(ROOT_MAP):
         pytest.skip("root cost map only ships in source checkouts")
diff --git a/tests/test_litellm/test_main.py b/tests/test_litellm/test_main.py
index 8cf878d05d9..7c2b9d0be05 100644
--- a/tests/test_litellm/test_main.py
+++ b/tests/test_litellm/test_main.py
@@ -3150,8 +3150,8 @@ def _stream_builder_logging_obj() -> LiteLLMLogging:
     return logging_obj
 
 
-def test_stream_chunk_builder_reports_streaming_usage_cost_when_enabled(monkeypatch: pytest.MonkeyPatch):
-    monkeypatch.setattr(litellm, "include_cost_in_streaming_usage", True)
+def test_stream_chunk_builder_stamps_streaming_usage_cost_by_default(monkeypatch: pytest.MonkeyPatch):
+    monkeypatch.setattr(litellm, "include_cost_in_streaming_usage", False)
     chunks: Final = [
         _stream_builder_text_chunk("gpt-4o", "Hello "),
         _stream_builder_text_chunk("gpt-4o", "world.", finish_reason="stop"),
@@ -3168,11 +3168,45 @@ def test_stream_chunk_builder_reports_streaming_usage_cost_when_enabled(monkeypa
     assert response._hidden_params["response_cost"] == pytest.approx(usage_cost)
 
 
-def test_stream_chunk_builder_defers_cost_to_logging_obj_when_usage_cost_absent(monkeypatch: pytest.MonkeyPatch):
-    monkeypatch.setattr(litellm, "include_cost_in_streaming_usage", False)
+def test_stream_chunk_builder_skips_stamp_when_cost_is_unpriceable():
+    import time as time_module
+
+    from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLogging
+
+    logging_obj: Final = LiteLLMLogging(
+        model="us.anthropic.claude-opus-5",
+        messages=[{"role": "user", "content": "hi"}],
+        stream=True,
+        call_type="completion",
+        start_time=time_module.time(),
+        litellm_call_id="stream-builder-alias-unpriceable",
+        function_id="1",
+    )
+    logging_obj.model_call_details["custom_llm_provider"] = "bedrock"
+    logging_obj.optional_params = {}
+    usage_chunk: Final = _stream_builder_text_chunk("bedrock-claude-opus-5", "")
+    usage_chunk.usage = Usage(prompt_tokens=40, completion_tokens=5, total_tokens=45)
+    chunks: Final = [
+        _stream_builder_text_chunk("bedrock-claude-opus-5", "Hello ", finish_reason="stop"),
+        usage_chunk,
+    ]
+
+    response: Final = litellm.stream_chunk_builder(
+        chunks=chunks, messages=[{"role": "user", "content": "hi"}], logging_obj=logging_obj
+    )
+
+    assert response is not None
+    assert getattr(response.usage, "cost", None) is None
+    assert response._hidden_params.get("response_cost") is None
+
+
+def test_stream_chunk_builder_keeps_provider_reported_usage_cost():
+    usage_chunk: Final = _stream_builder_text_chunk("gpt-4o", "")
+    usage_chunk.usage = Usage(prompt_tokens=10, completion_tokens=5, total_tokens=15, cost=0.5)
     chunks: Final = [
         _stream_builder_text_chunk("gpt-4o", "Hello "),
         _stream_builder_text_chunk("gpt-4o", "world.", finish_reason="stop"),
+        usage_chunk,
     ]
 
     response: Final = litellm.stream_chunk_builder(
@@ -3180,4 +3214,26 @@ def test_stream_chunk_builder_defers_cost_to_logging_obj_when_usage_cost_absent(
     )
 
     assert response is not None
-    assert response._hidden_params.get("response_cost") is None
+    assert getattr(response.usage, "cost", None) == pytest.approx(0.5)
+    assert response._hidden_params["response_cost"] == pytest.approx(0.5)
+
+
+def test_stream_chunk_builder_prices_alias_from_openai_sdk_usage_chunk():
+    from openai.types.completion_usage import CompletionUsage
+
+    usage_chunk: Final = _stream_builder_text_chunk("mantle-claude", "")
+    usage_chunk.usage = CompletionUsage(prompt_tokens=20, completion_tokens=60, total_tokens=80, cost=0.000704)
+    assert type(usage_chunk.usage) is CompletionUsage
+    chunks: Final = [
+        _stream_builder_text_chunk("mantle-claude", "Hello "),
+        _stream_builder_text_chunk("mantle-claude", "world.", finish_reason="stop"),
+        usage_chunk,
+    ]
+
+    response: Final = litellm.stream_chunk_builder(chunks=chunks, messages=[{"role": "user", "content": "hi"}])
+
+    assert response is not None
+    assert response.usage.prompt_tokens == 20
+    assert response.usage.completion_tokens == 60
+    assert getattr(response.usage, "cost", None) == pytest.approx(0.000704)
+    assert response._hidden_params["response_cost"] == pytest.approx(0.000704)
diff --git a/tests/test_litellm/test_openai_service_tier_long_context_pricing.py b/tests/test_litellm/test_openai_service_tier_long_context_pricing.py
new file mode 100644
index 00000000000..c0860a5b55f
--- /dev/null
+++ b/tests/test_litellm/test_openai_service_tier_long_context_pricing.py
@@ -0,0 +1,156 @@
+import json
+from functools import lru_cache
+from pathlib import Path
+
+import pytest
+
+import litellm
+
+REPO_ROOT = Path(__file__).parents[2]
+MAIN_PATH = REPO_ROOT / "model_prices_and_context_window.json"
+BACKUP_PATH = REPO_ROOT / "litellm" / "model_prices_and_context_window_backup.json"
+
+FLEX_LONG_CONTEXT = {
+    "gpt-5.4": {
+        "input_cost_per_token_above_272k_tokens_flex": 2.5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 1.125e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 2.5e-07,
+    },
+    "gpt-5.4-pro": {
+        "input_cost_per_token_above_272k_tokens_flex": 3e-05,
+        "output_cost_per_token_above_272k_tokens_flex": 0.000135,
+    },
+    "gpt-5.5": {
+        "input_cost_per_token_above_272k_tokens_flex": 5e-06,
+        "output_cost_per_token_above_272k_tokens_flex": 2.25e-05,
+        "cache_read_input_token_cost_above_272k_tokens_flex": 5e-07,
+    },
+}
+
+PRIORITY_LONG_CONTEXT = {
+    "gpt-5.6": {
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
+    },
+    "gpt-5.6-sol": {
+        "input_cost_per_token_above_272k_tokens_priority": 1.6e-05,
+        "output_cost_per_token_above_272k_tokens_priority": 6e-05,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 1.6e-06,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 2e-05,
+    },
+    "gpt-5.6-terra": {
+        "input_cost_per_token_above_272k_tokens_priority": 8e-06,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-05,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-07,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-05,
+    },
+    "gpt-5.6-luna": {
+        "input_cost_per_token_above_272k_tokens_priority": 8e-07,
+        "output_cost_per_token_above_272k_tokens_priority": 3.6e-06,
+        "cache_read_input_token_cost_above_272k_tokens_priority": 8e-08,
+        "cache_creation_input_token_cost_above_272k_tokens_priority": 1e-06,
+    },
+}
+
+EXPECTED = {**FLEX_LONG_CONTEXT, **PRIORITY_LONG_CONTEXT}
+
+NO_PUBLISHED_PRIORITY_LONG_CONTEXT = ("gpt-5.4", "gpt-5.5")
+
+
+@pytest.fixture(autouse=True)
+def _local_model_cost_map(monkeypatch: pytest.MonkeyPatch) -> None:
+    monkeypatch.setenv("LITELLM_LOCAL_MODEL_COST_MAP", "True")
+    monkeypatch.setattr(litellm, "model_cost", litellm.get_model_cost_map(url=""))
+
+
+@lru_cache(maxsize=2)
+def _load(path: Path) -> dict[str, dict[str, object]]:
+    with open(path) as f:
+        return json.load(f)
+
+
+@pytest.mark.parametrize("path", [MAIN_PATH, BACKUP_PATH], ids=["main", "backup"])
+@pytest.mark.parametrize("model", sorted(EXPECTED))
+def test_service_tier_long_context_rates_are_published(model: str, path: Path) -> None:
+    """Each tier must carry its own above-272K rates, in both price files."""
+    info = _load(path).get(model)
+    assert info is not None, f"{model} not found in {path.name}"
+    for key, expected in EXPECTED[model].items():
+        assert info.get(key) == pytest.approx(expected), f"{model}.{key} is {info.get(key)!r}, expected {expected!r}"
+
+
+@pytest.mark.parametrize("model", sorted(EXPECTED))
+def test_tier_long_context_rate_is_half_or_double_the_standard(model: str) -> None:
+    """Flex is half the standard long-context rate; priority is double it."""
+    info = _load(MAIN_PATH)[model]
+    tier = "flex" if model in FLEX_LONG_CONTEXT else "priority"
+    ratio = 0.5 if tier == "flex" else 2.0
+    for base in ("input_cost_per_token", "output_cost_per_token"):
+        standard = info[f"{base}_above_272k_tokens"]
+        tiered = info[f"{base}_above_272k_tokens_{tier}"]
+        assert tiered == pytest.approx(standard * ratio), (
+            f"{model}.{base}_above_272k_tokens_{tier} is {tiered!r}, "
+            f"expected {ratio}x the standard long-context rate {standard!r}"
+        )
+
+
+@pytest.mark.parametrize("model", NO_PUBLISHED_PRIORITY_LONG_CONTEXT)
+def test_no_priority_long_context_rates_where_openai_publishes_none(model: str) -> None:
+    """Guard against back-filling a rate OpenAI does not publish."""
+    info = _load(MAIN_PATH)[model]
+    assert "input_cost_per_token_above_272k_tokens_priority" not in info
+
+
+LONG_CONTEXT_PROMPT_TOKENS = 300_000
+COMPLETION_TOKENS = 1_000
+
+TIERED_COST_CASES = [
+    ("gpt-5.4", "flex", 2.5e-06, 1.125e-05),
+    ("gpt-5.4-pro", "flex", 3e-05, 0.000135),
+    ("gpt-5.5", "flex", 5e-06, 2.25e-05),
+    ("gpt-5.6", "priority", 1.6e-05, 6e-05),
+    ("gpt-5.6-sol", "priority", 1.6e-05, 6e-05),
+    ("gpt-5.6-terra", "priority", 8e-06, 3.6e-05),
+    ("gpt-5.6-luna", "priority", 8e-07, 3.6e-06),
+]
+
+
+@pytest.mark.parametrize("model,tier,input_rate,output_rate", TIERED_COST_CASES)
+def test_cost_per_token_bills_long_context_at_the_tier_rate(
+    model: str, tier: str, input_rate: float, output_rate: float
+) -> None:
+    """A prompt over 272K on flex or priority must bill at that tier's long-context rate."""
+    input_cost, output_cost = litellm.cost_per_token(
+        model=model,
+        prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
+        completion_tokens=COMPLETION_TOKENS,
+        service_tier=tier,
+    )
+    assert input_cost == pytest.approx(LONG_CONTEXT_PROMPT_TOKENS * input_rate)
+    assert output_cost == pytest.approx(COMPLETION_TOKENS * output_rate)
+
+
+@pytest.mark.parametrize("model,tier,input_rate,output_rate", TIERED_COST_CASES)
+def test_cost_per_token_tier_differs_from_the_standard_long_context_cost(
+    model: str, tier: str, input_rate: float, output_rate: float
+) -> None:
+    """Flex halves the standard long-context bill and priority doubles it."""
+    ratio = 0.5 if tier == "flex" else 2.0
+    standard = sum(
+        litellm.cost_per_token(
+            model=model,
+            prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
+            completion_tokens=COMPLETION_TOKENS,
+        )
+    )
+    tiered = sum(
+        litellm.cost_per_token(
+            model=model,
+            prompt_tokens=LONG_CONTEXT_PROMPT_TOKENS,
+            completion_tokens=COMPLETION_TOKENS,
+            service_tier=tier,
+        )
+    )
+    assert tiered == pytest.approx(standard * ratio)
diff --git a/tests/test_litellm/test_router.py b/tests/test_litellm/test_router.py
index 84f6344be35..ff5db18a3ea 100644
--- a/tests/test_litellm/test_router.py
+++ b/tests/test_litellm/test_router.py
@@ -7271,6 +7271,71 @@ def test_get_configured_token_limits_coerces_numeric_strings():
     assert router.get_configured_token_limits("quoted-limits-model") == (32000, 8000)
 
 
+def test_get_configured_display_name_reads_deployment_model_info():
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "Kimi K3-claude-compatible",
+                "litellm_params": {"model": "openai/some-unmapped-model"},
+                "model_info": {"display_name": "Kimi K3"},
+            }
+        ]
+    )
+
+    assert router.get_configured_display_name("Kimi K3-claude-compatible") == "Kimi K3"
+
+
+def test_get_configured_display_name_returns_none_for_unset_or_unknown():
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "no-display-model",
+                "litellm_params": {"model": "openai/some-unmapped-model"},
+            }
+        ]
+    )
+
+    assert router.get_configured_display_name("no-display-model") is None
+    assert router.get_configured_display_name("not-a-real-model") is None
+
+
+def test_get_configured_display_name_skips_wildcard_pattern_matching():
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "bedrock/*",
+                "litellm_params": {"model": "bedrock/*"},
+                "model_info": {"display_name": "Bedrock"},
+            }
+        ]
+    )
+
+    with patch.object(
+        router.pattern_router, "route", side_effect=AssertionError("pattern route called")
+    ):
+        assert (
+            router.get_configured_display_name("bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0")
+            is None
+        )
+
+
+def test_get_configured_display_name_treats_malformed_values_as_absent():
+    malformed = ["", "   ", 12345, ["Kimi K3"], {"name": "Kimi K3"}, True]
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": f"bad-display-{i}",
+                "litellm_params": {"model": "openai/some-unmapped-model"},
+                "model_info": {"display_name": bad},
+            }
+            for i, bad in enumerate(malformed)
+        ]
+    )
+
+    for i in range(len(malformed)):
+        assert router.get_configured_display_name(f"bad-display-{i}") is None
+
+
 @pytest.mark.asyncio
 async def test_acreate_batch_disable_fallbacks_surfaces_owning_provider_error():
     router = litellm.Router(
@@ -7510,6 +7575,118 @@ async def test_acreate_batch_request_bedrock_tags_override_deployment_tags():
         assert mock_sign.call_args.kwargs["data"]["tags"] == request_tags
 
 
+@pytest.mark.asyncio
+async def test_avector_store_search_injects_router():
+    """
+    Regression: router.avector_store_search must pass the router down to the
+    SDK search call so provider transforms can resolve router-managed
+    embedding models (e.g. S3 Vectors query embeddings).
+    """
+    from litellm.types.vector_stores import VectorStoreSearchResponse
+
+    expected_response = VectorStoreSearchResponse(
+        object="vector_store.search_results.page", search_query="q", data=[]
+    )
+    mock_asearch = AsyncMock(return_value=expected_response)
+    # Router.__init__ binds asearch via a local import, so patch the module
+    # attribute before constructing the Router.
+    with patch("litellm.vector_stores.main.asearch", new=mock_asearch):  # test-quality-ok: the SDK call is the only place the injected router kwarg is observable
+        router = litellm.Router(
+            model_list=[
+                {
+                    "model_name": "gpt-3.5-turbo",
+                    "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "test-key"},
+                }
+            ]
+        )
+        search_response = await router.avector_store_search(
+            vector_store_id="v", query="q", custom_llm_provider="s3_vectors"
+        )
+
+    assert search_response is expected_response
+    mock_asearch.assert_awaited_once()
+    assert mock_asearch.await_args.kwargs["router"] is router
+
+
+@pytest.mark.asyncio
+async def test_avector_store_create_does_not_inject_router():
+    """The router injection is gated on the search call type: the create path
+    must keep calling the SDK without a router kwarg."""
+    expected_response = {"id": "vs_1", "object": "vector_store"}
+    mock_acreate = AsyncMock(return_value=expected_response)
+    # avector_store_create(model=None) resolves acreate via a local import at
+    # call time, so patching after Router construction works here.
+    router = litellm.Router(
+        model_list=[
+            {
+                "model_name": "gpt-3.5-turbo",
+                "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "test-key"},
+            }
+        ]
+    )
+    with patch("litellm.vector_stores.main.acreate", new=mock_acreate):  # test-quality-ok: the SDK call is the only place a leaked router kwarg would surface
+        create_response = await router.avector_store_create(model=None, custom_llm_provider="openai")
+
+    assert create_response is expected_response
+    mock_acreate.assert_awaited_once()
+    assert "router" not in mock_acreate.await_args.kwargs
+
+
+def test_vector_store_search_injects_router():
+    """
+    Sync parity for the router injection: router.vector_store_search must pass
+    the router down to the SDK search call so provider transforms can resolve
+    router-managed embedding models, same as avector_store_search.
+    """
+    from litellm.types.vector_stores import VectorStoreSearchResponse
+
+    expected_response = VectorStoreSearchResponse(
+        object="vector_store.search_results.page", search_query="q", data=[]
+    )
+    mock_search = MagicMock(return_value=expected_response)
+    # Router.__init__ binds search via a local import, so patch the module
+    # attribute before constructing the Router.
+    with patch("litellm.vector_stores.main.search", new=mock_search):  # test-quality-ok: the SDK call is the only place the injected router kwarg is observable
+        router = litellm.Router(
+            model_list=[
+                {
+                    "model_name": "gpt-3.5-turbo",
+                    "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "test-key"},
+                }
+            ]
+        )
+        search_response = router.vector_store_search(
+            vector_store_id="v", query="q", custom_llm_provider="s3_vectors"
+        )
+
+    assert search_response is expected_response
+    mock_search.assert_called_once()
+    assert mock_search.call_args.kwargs["router"] is router
+    assert mock_search.call_args.kwargs["custom_llm_provider"] == "s3_vectors"
+
+
+def test_vector_store_create_does_not_inject_router():
+    """The sync create path must keep calling the SDK without a router kwarg."""
+    expected_response = {"id": "vs_1", "object": "vector_store"}
+    mock_create = MagicMock(return_value=expected_response)
+    # Router.__init__ binds create via a local import, so patch the module
+    # attribute before constructing the Router.
+    with patch("litellm.vector_stores.main.create", new=mock_create):  # test-quality-ok: the SDK call is the only place a leaked router kwarg would surface
+        router = litellm.Router(
+            model_list=[
+                {
+                    "model_name": "gpt-3.5-turbo",
+                    "litellm_params": {"model": "openai/gpt-3.5-turbo", "api_key": "test-key"},
+                }
+            ]
+        )
+        create_response = router.vector_store_create(custom_llm_provider="openai")
+
+    assert create_response is expected_response
+    mock_create.assert_called_once()
+    assert "router" not in mock_create.call_args.kwargs
+
+
 class TestPreRoutingStrategyRegistryLifecycle:
     """
     Regression tests: a deployment leaving the model_list must release the
diff --git a/tests/test_litellm/test_router_retry_policy_update.py b/tests/test_litellm/test_router_retry_policy_update.py
index 1b98b8c1ae8..be568134763 100644
--- a/tests/test_litellm/test_router_retry_policy_update.py
+++ b/tests/test_litellm/test_router_retry_policy_update.py
@@ -21,6 +21,7 @@ This file pins both halves of the fix.
 
 import json
 from dataclasses import dataclass
+from typing import Final
 from unittest.mock import AsyncMock, MagicMock
 
 import pytest
@@ -28,8 +29,19 @@ from pydantic import ValidationError
 
 
 import litellm
+from litellm.router_strategy.budget_limiter import RouterBudgetLimiting
+from litellm.router_utils.pre_call_checks.model_rate_limit_check import ModelRateLimitingCheck
+from litellm.router_utils.pre_call_checks.prompt_caching_deployment_check import PromptCachingDeploymentCheck
 from litellm.types.router import RetryPolicy, UpdateRouterConfig
 
+
+@pytest.fixture(autouse=True)
+def isolate_litellm_callbacks():
+    callbacks_before: Final = litellm.callbacks.copy()
+    yield
+    litellm.callbacks = callbacks_before  # test-quality-ok: required callback-state restoration fixture
+
+
 # ---------------------------------------------------------------------------
 # UpdateRouterConfig schema membership (LIT-3152 part 1)
 # ---------------------------------------------------------------------------
@@ -100,6 +112,114 @@ def _build_router() -> litellm.Router:
     )
 
 
+def test_update_settings_adds_optional_pre_call_check_once():
+    router = _build_router()
+
+    router.update_settings(num_retries=7, optional_pre_call_checks=["prompt_caching"])
+    router.update_settings(optional_pre_call_checks=["prompt_caching"])
+
+    prompt_caching_callbacks = [
+        callback for callback in router.optional_callbacks if isinstance(callback, PromptCachingDeploymentCheck)
+    ]
+    assert len(prompt_caching_callbacks) == 1
+    assert router.num_retries == 7
+
+
+def test_update_settings_clears_omitted_toggleable_pre_call_checks():
+    router = _build_router()
+
+    router.update_settings(optional_pre_call_checks=["prompt_caching"])
+    router.update_settings(optional_pre_call_checks=[])
+
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in (router.optional_callbacks or []))
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in litellm.callbacks)
+
+
+def test_set_optional_pre_call_checks_reconciles_callback_types():
+    router = _build_router()
+
+    router.set_optional_pre_call_checks(["prompt_caching"])
+    router.set_optional_pre_call_checks([])
+
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in (router.optional_callbacks or []))
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in litellm.callbacks)
+
+
+def test_remove_optional_pre_call_check_removes_local_and_global_callbacks():
+    router = _build_router()
+
+    router.set_optional_pre_call_checks(["prompt_caching"])
+    router._remove_optional_callbacks_of_type(PromptCachingDeploymentCheck)
+
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in (router.optional_callbacks or []))
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in litellm.callbacks)
+
+
+def test_remove_optional_pre_call_check_keeps_global_callback_for_another_router():
+    router_a = _build_router()
+    router_b = _build_router()
+
+    router_a.update_settings(optional_pre_call_checks=["prompt_caching"])
+    router_b.update_settings(optional_pre_call_checks=["prompt_caching"])
+
+    router_a.update_settings(optional_pre_call_checks=[])
+
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in (router_a.optional_callbacks or []))
+    assert any(type(callback) is PromptCachingDeploymentCheck for callback in (router_b.optional_callbacks or []))
+    assert any(type(callback) is PromptCachingDeploymentCheck for callback in litellm.callbacks)
+
+    router_b.update_settings(optional_pre_call_checks=[])
+
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in (router_b.optional_callbacks or []))
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in litellm.callbacks)
+
+
+def test_remove_optional_pre_call_check_keeps_global_callback_when_second_router_clears_first():
+    router_a = _build_router()
+    router_b = _build_router()
+
+    router_a.update_settings(optional_pre_call_checks=["prompt_caching"])
+    router_b.update_settings(optional_pre_call_checks=["prompt_caching"])
+
+    router_b.update_settings(optional_pre_call_checks=[])
+
+    assert any(type(callback) is PromptCachingDeploymentCheck for callback in (router_a.optional_callbacks or []))
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in (router_b.optional_callbacks or []))
+    assert any(type(callback) is PromptCachingDeploymentCheck for callback in litellm.callbacks)
+
+    router_a.update_settings(optional_pre_call_checks=[])
+
+    assert not any(type(callback) is PromptCachingDeploymentCheck for callback in litellm.callbacks)
+
+
+def test_update_settings_replaces_toggleable_pre_call_checks():
+    router = _build_router()
+
+    router.update_settings(optional_pre_call_checks=["prompt_caching"])
+    router.update_settings(optional_pre_call_checks=["enforce_model_rate_limits"])
+
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in (router.optional_callbacks or []))
+    assert not any(isinstance(callback, PromptCachingDeploymentCheck) for callback in litellm.callbacks)
+    assert any(isinstance(callback, ModelRateLimitingCheck) for callback in (router.optional_callbacks or []))
+
+
+@pytest.mark.asyncio
+async def test_update_settings_preserves_router_budget_limiting_when_omitted(monkeypatch):
+    async def _disable_periodic_sync(*args, **kwargs):
+        return None
+
+    monkeypatch.setattr(
+        "litellm.router_strategy.budget_limiter.RouterBudgetLimiting.periodic_sync_in_memory_spend_with_redis",
+        _disable_periodic_sync,
+    )
+    router = _build_router()
+
+    router.add_optional_pre_call_checks(["router_budget_limiting"])
+    router.update_settings(optional_pre_call_checks=[])
+
+    assert any(isinstance(callback, RouterBudgetLimiting) for callback in (router.optional_callbacks or []))
+
+
 def test_update_settings_persists_retry_policy_dict():
     """When the proxy's ``_add_router_settings_from_db_config`` calls
     ``llm_router.update_settings(retry_policy={...})`` after reading the
@@ -255,8 +375,12 @@ async def test_config_update_persists_and_reads_back_retry_policy(monkeypatch):
             RateLimitErrorRetries=7,
         )
     )
+    request = MagicMock()
+    request.json = AsyncMock(return_value={"router_settings": {"retry_policy": posted.model_dump()}})
+
     await proxy_server.update_config(
         config_info=ConfigYAML(router_settings=posted),
+        request=request,
         user_api_key_dict=UserAPIKeyAuth(user_role=LitellmUserRoles.PROXY_ADMIN, api_key="sk-1234"),
     )
 
diff --git a/tests/test_litellm/test_utils.py b/tests/test_litellm/test_utils.py
index 521e91daded..0790b41c349 100644
--- a/tests/test_litellm/test_utils.py
+++ b/tests/test_litellm/test_utils.py
@@ -4655,6 +4655,7 @@ GEMINI_4096_CACHE_MIN_MODELS: Final = tuple(
         "gemini-3.5-flash",
         "gemini-3.6-flash",
         "gemini-3.7-flash",
+        "gemini-3.8-flash",
         "gemini-3.1-pro-preview",
         "gemini-3.1-pro-preview-customtools",
     )
diff --git a/tests/test_litellm/vector_stores/__init__.py b/tests/test_litellm/vector_stores/__init__.py
new file mode 100644
index 00000000000..e69de29bb2d
diff --git a/tests/test_litellm/vector_stores/test_main.py b/tests/test_litellm/vector_stores/test_main.py
new file mode 100644
index 00000000000..d01e696906a
--- /dev/null
+++ b/tests/test_litellm/vector_stores/test_main.py
@@ -0,0 +1,78 @@
+"""
+Tests for litellm/vector_stores/main.py.
+
+Pins the router threading contract for vector store search: the router is an
+explicit named parameter that reaches the HTTP handler, and it must never leak
+into litellm_params/kwargs where logging would model_dump() it (the #19550
+serialization trap).
+"""
+
+from unittest.mock import MagicMock, patch
+
+import litellm.vector_stores.main as vector_stores_main
+from litellm.vector_stores.main import search
+
+MOCK_SEARCH_RESPONSE = {
+    "object": "vector_store.search_results.page",
+    "search_query": "q",
+    "data": [],
+}
+
+
+def test_search_threads_router_to_handler():
+    """search() must pass its router param through to the HTTP handler"""
+    mock_router = MagicMock()
+    logger = MagicMock()
+
+    with (
+        patch(  # test-quality-ok: stubs provider config resolution; the seam under test is the router kwarg threading
+            "litellm.vector_stores.main.ProviderConfigManager.get_provider_vector_stores_config",
+            return_value=MagicMock(),
+        ),
+        patch.object(  # test-quality-ok: the handler call is the observable boundary for the router kwarg contract
+            vector_stores_main.base_llm_http_handler,
+            "vector_store_search_handler",
+            return_value=MOCK_SEARCH_RESPONSE,
+        ) as mock_handler,
+    ):
+        response = search(
+            vector_store_id="bkt:idx",
+            query="q",
+            custom_llm_provider="s3_vectors",
+            router=mock_router,
+            litellm_logging_obj=logger,
+        )
+
+    assert response == MOCK_SEARCH_RESPONSE
+    mock_handler.assert_called_once()
+    assert mock_handler.call_args.kwargs["router"] is mock_router
+
+
+def test_search_router_not_in_litellm_params():
+    """Regression (#19550 class): the router must stay out of GenericLiteLLMParams,
+    otherwise pre-call logging model_dump()s it and breaks serialization."""
+    mock_router = MagicMock()
+    logger = MagicMock()
+
+    with (
+        patch(  # test-quality-ok: stubs provider config resolution; the seam under test is litellm_params contents
+            "litellm.vector_stores.main.ProviderConfigManager.get_provider_vector_stores_config",
+            return_value=MagicMock(),
+        ),
+        patch.object(  # test-quality-ok: the handler call is where a leaked router in litellm_params would surface
+            vector_stores_main.base_llm_http_handler,
+            "vector_store_search_handler",
+            return_value=MOCK_SEARCH_RESPONSE,
+        ) as mock_handler,
+    ):
+        search(
+            vector_store_id="bkt:idx",
+            query="q",
+            custom_llm_provider="s3_vectors",
+            router=mock_router,
+            litellm_logging_obj=logger,
+        )
+
+    litellm_params = mock_handler.call_args.kwargs["litellm_params"]
+    assert "router" not in litellm_params.model_dump(exclude_none=True)
+    assert getattr(litellm_params, "router", None) is None
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_info.integration.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_info.integration.test.tsx
index 5bc8c2a5b1c..357f924cbe7 100644
--- a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_info.integration.test.tsx
+++ b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_info.integration.test.tsx
@@ -82,6 +82,40 @@ const langgraphInfo: AgentCreateInfo = {
   ],
 };
 
+const FULL_RUNTIME_ARN = "arn:aws:bedrock-agentcore:eu-central-1:123456789012:runtime/hosted_agent_4vm3i-BaTdfOELAs";
+
+const BEDROCK_AGENTCORE_AGENT = {
+  agent_id: "agent-3",
+  agent_name: "bedrock-agent",
+  agent_card_params: { name: "bedrock-agent", description: "agentcore agent", url: "", version: "1.0.0", skills: [] },
+  litellm_params: {
+    custom_llm_provider: "bedrock",
+    model: `bedrock/agentcore/${FULL_RUNTIME_ARN}`,
+  },
+};
+
+const bedrockAgentcoreInfo: AgentCreateInfo = {
+  agent_type: "bedrock_agentcore",
+  agent_type_display_name: "Bedrock AgentCore",
+  description: "Bedrock AgentCore runtimes",
+  logo_url: "/b.png",
+  use_a2a_form_fields: false,
+  litellm_params_template: { custom_llm_provider: "bedrock" },
+  model_template: "bedrock/agentcore/{agent_runtime_arn}",
+  credential_fields: [
+    {
+      key: "agent_runtime_arn",
+      label: "Agent Runtime ARN",
+      field_type: "text",
+      required: true,
+      include_in_litellm_params: false,
+      validation_pattern: "^arn:aws[a-zA-Z0-9-]*:bedrock-agentcore:[a-z0-9-]+:[0-9]{12}:runtime/.+$",
+      validation_message:
+        'Enter the complete Bedrock AgentCore runtime ARN, including the runtime ID after "runtime/".',
+    },
+  ],
+};
+
 const setup = () => userEvent.setup({ pointerEventsCheck: PointerEventsCheckLevel.Never });
 
 const renderView = () => {
@@ -284,6 +318,69 @@ describe("AgentInfoView update payload", () => {
     expect(patchedPayload().object_permission).toEqual(existingMcpGrants);
   });
 
+  it("preserves the full AgentCore runtime ARN (including the resource id after runtime/) across an unedited save", async () => {
+    vi.mocked(networking.getAgentCreateMetadata).mockResolvedValue([bedrockAgentcoreInfo]);
+    vi.mocked(networking.getAgentInfo).mockResolvedValue(BEDROCK_AGENTCORE_AGENT as never);
+    const user = setup();
+    renderView();
+    await openEditor(user);
+
+    expect(await screen.findByLabelText("Agent Runtime ARN")).toHaveValue(FULL_RUNTIME_ARN);
+
+    await save(user);
+
+    expect((patchedPayload().litellm_params as Record<string, unknown>).model).toBe(
+      `bedrock/agentcore/${FULL_RUNTIME_ARN}`,
+    );
+  });
+
+  it("blocks the save and shows a validation error when the Agent Runtime ARN is truncated", async () => {
+    vi.mocked(networking.getAgentCreateMetadata).mockResolvedValue([bedrockAgentcoreInfo]);
+    vi.mocked(networking.getAgentInfo).mockResolvedValue(BEDROCK_AGENTCORE_AGENT as never);
+    const user = setup();
+    renderView();
+    await openEditor(user);
+
+    const arnField = await screen.findByLabelText("Agent Runtime ARN");
+    await user.clear(arnField);
+    fireEvent.change(arnField, {
+      target: { value: "arn:aws:bedrock-agentcore:eu-central-1:123456789012:runtime" },
+    });
+    await user.click(screen.getByRole("button", { name: "Save Changes" }));
+
+    expect(
+      await screen.findByText(
+        'Enter the complete Bedrock AgentCore runtime ARN, including the runtime ID after "runtime/".',
+      ),
+    ).toBeInTheDocument();
+    expect(networking.patchAgentCall).not.toHaveBeenCalled();
+  });
+
+  it("renders and saves normally when a field's validation_pattern is not a valid regex", async () => {
+    const infoWithBadPattern: AgentCreateInfo = {
+      ...bedrockAgentcoreInfo,
+      credential_fields: [
+        {
+          ...bedrockAgentcoreInfo.credential_fields[0],
+          validation_pattern: "(unterminated",
+        },
+      ],
+    };
+    vi.mocked(networking.getAgentCreateMetadata).mockResolvedValue([infoWithBadPattern]);
+    vi.mocked(networking.getAgentInfo).mockResolvedValue(BEDROCK_AGENTCORE_AGENT as never);
+    const user = setup();
+    renderView();
+    await openEditor(user);
+
+    expect(await screen.findByLabelText("Agent Runtime ARN")).toHaveValue(FULL_RUNTIME_ARN);
+
+    await save(user);
+
+    expect((patchedPayload().litellm_params as Record<string, unknown>).model).toBe(
+      `bedrock/agentcore/${FULL_RUNTIME_ARN}`,
+    );
+  });
+
   it("reloads the agent and leaves edit mode when the edit is cancelled", async () => {
     const user = setup();
     renderView();
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.test.ts b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.test.ts
new file mode 100644
index 00000000000..0c2d2500776
--- /dev/null
+++ b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.test.ts
@@ -0,0 +1,73 @@
+import { describe, it, expect } from "vitest";
+import { detectAgentType, extractModelTemplateValues, parseDynamicAgentForForm } from "./agent_type_utils";
+import type { AgentCreateInfo } from "@/components/networking";
+import type { Agent } from "@/components/agents/types";
+
+const FULL_RUNTIME_ARN = "arn:aws:bedrock-agentcore:eu-central-1:123456789012:runtime/hosted_agent_4vm3i-BaTdfOELAs";
+
+const bedrockAgentcoreInfo: AgentCreateInfo = {
+  agent_type: "bedrock_agentcore",
+  agent_type_display_name: "Bedrock AgentCore",
+  model_template: "bedrock/agentcore/{agent_runtime_arn}",
+  credential_fields: [
+    {
+      key: "agent_runtime_arn",
+      label: "Agent Runtime ARN",
+      required: true,
+      include_in_litellm_params: false,
+    },
+  ],
+};
+
+describe("extractModelTemplateValues", () => {
+  it("recovers a placeholder value that itself contains '/' (an AWS ARN resource path)", () => {
+    const values = extractModelTemplateValues(
+      "bedrock/agentcore/{agent_runtime_arn}",
+      `bedrock/agentcore/${FULL_RUNTIME_ARN}`,
+    );
+
+    expect(values.agent_runtime_arn).toBe(FULL_RUNTIME_ARN);
+  });
+
+  it("recovers a placeholder value with no '/' (single path segment)", () => {
+    const values = extractModelTemplateValues("langgraph/{assistant_id}", "langgraph/asst_1");
+
+    expect(values.assistant_id).toBe("asst_1");
+  });
+
+  it("returns no match when the model does not fit the template", () => {
+    const values = extractModelTemplateValues("langgraph/{assistant_id}", "azure_ai/agents/asst_1");
+
+    expect(values).toEqual({});
+  });
+});
+
+describe("parseDynamicAgentForForm", () => {
+  it("preserves the full runtime ARN, including the resource id after 'runtime/', when populating the edit form", () => {
+    const agent = {
+      agent_id: "agent-1",
+      agent_name: "bedrock-agent",
+      agent_card_params: { description: "" },
+      litellm_params: {
+        custom_llm_provider: "bedrock",
+        model: `bedrock/agentcore/${FULL_RUNTIME_ARN}`,
+      },
+    } as unknown as Agent;
+
+    const values = parseDynamicAgentForForm(agent, bedrockAgentcoreInfo);
+
+    expect(values.agent_runtime_arn).toBe(FULL_RUNTIME_ARN);
+  });
+});
+
+describe("detectAgentType", () => {
+  it("detects bedrock_agentcore agents from the model prefix", () => {
+    const agent = {
+      agent_id: "agent-1",
+      agent_name: "bedrock-agent",
+      litellm_params: { model: `bedrock/agentcore/${FULL_RUNTIME_ARN}` },
+    } as unknown as Agent;
+
+    expect(detectAgentType(agent)).toBe("bedrock_agentcore");
+  });
+});
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.ts b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.ts
index f91590c5732..506f355c519 100644
--- a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.ts
+++ b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/agent_type_utils.ts
@@ -25,6 +25,29 @@ export const detectAgentType = (agent: Agent): string => {
   return "a2a";
 };
 
+const escapeRegExp = (segment: string): string => segment.replace(/[.*+?^${}()|[\]\\]/g, "\\$&");
+
+/**
+ * Reverses a `model_template` (e.g. "bedrock/agentcore/{agent_runtime_arn}") against a stored
+ * `model` string to recover the placeholder values that produced it. Builds a regex from the
+ * template's literal segments rather than matching by split("/") position, because a
+ * placeholder's value can itself contain "/" (an AWS ARN's "runtime/<runtime-id>" resource path,
+ * a Vertex AI reasoning engine's "projects/.../reasoningEngines/..." resource id) and would
+ * otherwise be cut off at the first one.
+ */
+export const extractModelTemplateValues = (template: string, model: string): Record<string, string> => {
+  // Splitting on a regex with a capturing group interleaves the captured placeholder
+  // names between the surrounding literal segments, e.g. "a/{x}/b" -> ["a/", "x", "/b"].
+  const parts = template.split(/\{([a-zA-Z0-9_]+)\}/g);
+  const fieldNames = parts.filter((_part, index) => index % 2 === 1);
+  const pattern = parts.map((part, index) => (index % 2 === 1 ? "(.+)" : escapeRegExp(part))).join("");
+
+  const match = model.match(new RegExp(`^${pattern}$`));
+  if (!match) return {};
+
+  return Object.fromEntries(fieldNames.map((name, index) => [name, match[index + 1]]));
+};
+
 /**
  * Parses agent data for dynamic form fields (non-A2A agents).
  * Extracts values from litellm_params based on the agent type metadata.
@@ -35,24 +58,18 @@ export const parseDynamicAgentForForm = (agent: Agent, agentTypeInfo: AgentCreat
     description: agent.agent_card_params?.description || "",
   };
 
+  const templateValues =
+    agentTypeInfo.model_template && agent.litellm_params?.model
+      ? extractModelTemplateValues(agentTypeInfo.model_template, agent.litellm_params.model)
+      : {};
+
   // Extract credential field values from litellm_params
   for (const field of agentTypeInfo.credential_fields) {
     if (field.include_in_litellm_params !== false) {
       values[field.key] = agent.litellm_params?.[field.key] || field.default_value || "";
-    } else {
-      // For fields not in litellm_params (like agent_id), try to extract from model string
-      if (agentTypeInfo.model_template && agent.litellm_params?.model) {
-        const model = agent.litellm_params.model;
-        const templateParts = agentTypeInfo.model_template.split("/");
-        const modelParts = model.split("/");
-
-        // Find the placeholder position and extract the value
-        templateParts.forEach((part, index) => {
-          if (part === `{${field.key}}` && modelParts[index]) {
-            values[field.key] = modelParts[index];
-          }
-        });
-      }
+    } else if (templateValues[field.key] !== undefined) {
+      // For fields not in litellm_params (like agent_runtime_arn), recover from the model string
+      values[field.key] = templateValues[field.key];
     }
   }
 
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx
index 04a8b0df9d9..0f3355fe073 100644
--- a/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx
+++ b/ui/litellm-dashboard/src/app/(dashboard)/agents/_components/dynamic_agent_form_fields.tsx
@@ -24,58 +24,80 @@ interface DynamicAgentFormFieldsProps {
 export const unmountedDynamicFieldNames = (mountedPanels: readonly string[]): readonly string[] =>
   mountedPanels.includes(AGENT_FORM_CONFIG.cost.key) ? [] : COST_FIELD_NAMES;
 
-const CredentialField = ({ field }: { field: AgentCredentialFieldMetadata }) => (
-  <AgentFormField
-    name={field.key}
-    label={field.tooltip ? labelWithHint(field.label, field.tooltip) : field.label}
-    defaultValue={field.default_value ?? undefined}
-    rules={field.required ? { required: `Please enter ${field.label}` } : undefined}
-  >
-    {({ value, onChange, ref, ...control }) => {
-      const text = typeof value === "string" ? value : "";
-      if (field.field_type === "password") {
-        return (
-          <PasswordInput
-            {...control}
-            value={typeof value === "string" ? value : ""}
-            onChange={onChange}
-            ref={ref}
-            placeholder={field.placeholder || ""}
-          />
-        );
-      }
-      if (field.field_type === "textarea") {
-        return (
-          <Textarea
-            {...control}
-            ref={ref}
-            rows={3}
-            placeholder={field.placeholder || ""}
-            value={text}
-            onChange={onChange}
-          />
-        );
-      }
-      if (field.field_type === "select" && field.options) {
-        return (
-          <Select value={text || null} onValueChange={onChange}>
-            <SelectTrigger {...control} className="w-full">
-              <SelectValue placeholder={field.placeholder || ""} />
-            </SelectTrigger>
-            <SelectContent>
-              {field.options.map((option) => (
-                <SelectItem key={option} value={option} title={option}>
-                  {option}
-                </SelectItem>
-              ))}
-            </SelectContent>
-          </Select>
-        );
-      }
-      return <Input {...control} ref={ref} placeholder={field.placeholder || ""} value={text} onChange={onChange} />;
-    }}
-  </AgentFormField>
-);
+// A field's validation_pattern is server-supplied metadata; if it's ever not a valid regex, skip
+// validation rather than throwing during render and taking the whole form down with it.
+const buildValidationPatternRule = (
+  field: AgentCredentialFieldMetadata,
+): { value: RegExp; message: string } | undefined => {
+  if (!field.validation_pattern) return undefined;
+  try {
+    return {
+      value: new RegExp(field.validation_pattern),
+      message: field.validation_message || `${field.label} looks incomplete or malformed`,
+    };
+  } catch {
+    return undefined;
+  }
+};
+
+const CredentialField = ({ field }: { field: AgentCredentialFieldMetadata }) => {
+  const patternRule = buildValidationPatternRule(field);
+  return (
+    <AgentFormField
+      name={field.key}
+      label={field.tooltip ? labelWithHint(field.label, field.tooltip) : field.label}
+      defaultValue={field.default_value ?? undefined}
+      rules={{
+        ...(field.required ? { required: `Please enter ${field.label}` } : {}),
+        ...(patternRule ? { pattern: patternRule } : {}),
+      }}
+    >
+      {({ value, onChange, ref, ...control }) => {
+        const text = typeof value === "string" ? value : "";
+        if (field.field_type === "password") {
+          return (
+            <PasswordInput
+              {...control}
+              value={typeof value === "string" ? value : ""}
+              onChange={onChange}
+              ref={ref}
+              placeholder={field.placeholder || ""}
+            />
+          );
+        }
+        if (field.field_type === "textarea") {
+          return (
+            <Textarea
+              {...control}
+              ref={ref}
+              rows={3}
+              placeholder={field.placeholder || ""}
+              value={text}
+              onChange={onChange}
+            />
+          );
+        }
+        if (field.field_type === "select" && field.options) {
+          return (
+            <Select value={text || null} onValueChange={onChange}>
+              <SelectTrigger {...control} className="w-full">
+                <SelectValue placeholder={field.placeholder || ""} />
+              </SelectTrigger>
+              <SelectContent>
+                {field.options.map((option) => (
+                  <SelectItem key={option} value={option} title={option}>
+                    {option}
+                  </SelectItem>
+                ))}
+              </SelectContent>
+            </Select>
+          );
+        }
+        return <Input {...control} ref={ref} placeholder={field.placeholder || ""} value={text} onChange={onChange} />;
+      }}
+    </AgentFormField>
+  );
+};
 
 const DynamicAgentFormFields: React.FC<DynamicAgentFormFieldsProps> = ({ agentTypeInfo, panels }) => (
   <>
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.test.tsx b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.test.tsx
index 34643811e29..4203a3dbf70 100644
--- a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.test.tsx
+++ b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.test.tsx
@@ -116,16 +116,33 @@ describe("VectorStoreTester", () => {
     await waitFor(() => expect(mockSearch).toHaveBeenCalledTimes(1));
   });
 
-  it("reports a failed search and keeps the history empty", async () => {
+  it("shows the backend error in the history when a search fails", async () => {
     const user = userEvent.setup();
-    mockSearch.mockRejectedValue(new Error("boom"));
+    const errorBody = '{"error":{"message":"OpenAIException - api_key is required"}}';
+    mockSearch.mockRejectedValue(new Error(errorBody));
     renderTester();
 
     await user.type(queryInput(), "hello");
     await user.click(searchButton());
 
-    await waitFor(() => expect(mockFromBackend).toHaveBeenCalledWith("Failed to search vector store"));
-    expect(screen.getByText(EMPTY_STATE)).toBeInTheDocument();
+    await waitFor(() => expect(mockFromBackend).toHaveBeenCalledWith(errorBody));
+    expect(screen.getByText(`Search failed: ${errorBody}`)).toBeInTheDocument();
+    expect(screen.queryByText("No results found")).not.toBeInTheDocument();
+    expect(screen.queryByText(EMPTY_STATE)).not.toBeInTheDocument();
+    // the failed query stays in the input for retry
+    expect(queryInput()).toHaveValue("hello");
+  });
+
+  it('renders "No results found" for an empty result set, not an error', async () => {
+    const user = userEvent.setup();
+    mockSearch.mockResolvedValue({ object: "vector_store.search_results.page", search_query: "hello", data: [] });
+    renderTester();
+
+    await user.type(queryInput(), "hello");
+    await user.click(searchButton());
+
+    expect(await screen.findByText("No results found")).toBeInTheDocument();
+    expect(screen.queryByText(/search failed/i)).not.toBeInTheDocument();
   });
 
   it("clears the search history", async () => {
diff --git a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.tsx b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.tsx
index 6d12880319b..a9c4e0d1061 100644
--- a/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.tsx
+++ b/ui/litellm-dashboard/src/app/(dashboard)/vector-stores/_components/VectorStoreTester.tsx
@@ -40,6 +40,7 @@ export const VectorStoreTester: React.FC<VectorStoreTesterProps> = ({ vectorStor
     {
       query: string;
       response: VectorStoreSearchResponse | null;
+      error: string | null;
       timestamp: number;
     }[]
   >([]);
@@ -59,6 +60,7 @@ export const VectorStoreTester: React.FC<VectorStoreTesterProps> = ({ vectorStor
       const historyEntry = {
         query,
         response,
+        error: null,
         timestamp: Date.now(),
       };
 
@@ -66,7 +68,9 @@ export const VectorStoreTester: React.FC<VectorStoreTesterProps> = ({ vectorStor
       setQuery("");
     } catch (error) {
       console.error("Error searching vector store:", error);
-      toast.fromError("Failed to search vector store");
+      const errorMessage = error instanceof Error ? error.message : String(error);
+      toast.fromError(errorMessage);
+      setSearchHistory((prev) => [{ query, response: null, error: errorMessage, timestamp: Date.now() }, ...prev]);
     } finally {
       setIsLoading(false);
     }
@@ -228,7 +232,13 @@ export const VectorStoreTester: React.FC<VectorStoreTesterProps> = ({ vectorStor
                           })}
                         </div>
                       ) : (
-                        <div className="text-sm text-muted-foreground">No results found</div>
+                        <div
+                          className={
+                            entry.error ? "text-sm break-words text-destructive" : "text-sm text-muted-foreground"
+                          }
+                        >
+                          {entry.error ? `Search failed: ${entry.error}` : "No results found"}
+                        </div>
                       )}
                     </div>
                   </div>
diff --git a/ui/litellm-dashboard/src/components/networking.tsx b/ui/litellm-dashboard/src/components/networking.tsx
index 9c9458c81f2..6e45e4acdb1 100644
--- a/ui/litellm-dashboard/src/components/networking.tsx
+++ b/ui/litellm-dashboard/src/components/networking.tsx
@@ -310,6 +310,8 @@ export interface AgentCredentialFieldMetadata {
   options?: string[] | null;
   default_value?: string | null;
   include_in_litellm_params?: boolean;
+  validation_pattern?: string | null;
+  validation_message?: string | null;
 }
 
 export interface AgentCreateInfo {
@@ -6971,7 +6973,7 @@ export const vectorStoreSearchCall = async (
     if (!response.ok) {
       const errorData = await response.text();
       await handleError(errorData);
-      return null;
+      throw new Error(errorData);
     }
 
     const data = await response.json();
diff --git a/ui/litellm-dashboard/src/lib/http/schema.d.ts b/ui/litellm-dashboard/src/lib/http/schema.d.ts
index beeb574a3d9..7cb941dbf9b 100644
--- a/ui/litellm-dashboard/src/lib/http/schema.d.ts
+++ b/ui/litellm-dashboard/src/lib/http/schema.d.ts
@@ -22868,6 +22868,10 @@ export interface components {
             required: boolean;
             /** Tooltip */
             tooltip?: string | null;
+            /** Validation Message */
+            validation_message?: string | null;
+            /** Validation Pattern */
+            validation_pattern?: string | null;
         };
         /**
          * AgentExtension
@@ -37475,6 +37479,8 @@ export interface components {
             } | null;
             /** Num Retries */
             num_retries?: number | null;
+            /** Optional Pre Call Checks */
+            optional_pre_call_checks?: ("prompt_caching" | "router_budget_limiting" | "responses_api_deployment_check" | "deployment_affinity" | "session_affinity" | "forward_client_headers_by_model_group" | "enforce_model_rate_limits" | "encrypted_content_affinity")[] | null;
             /** Retry After */
             retry_after?: number | null;
             retry_policy?: components["schemas"]["RetryPolicy"] | null;
diff --git a/uv.lock b/uv.lock
index 27be919eea1..aa59ff7b229 100644
--- a/uv.lock
+++ b/uv.lock
@@ -10,7 +10,7 @@ resolution-markers = [
 ]
 
 [options]
-exclude-newer = "2026-08-29T17:58:57.633306Z"
+exclude-newer = "2026-08-30T17:51:25.171404Z"
 exclude-newer-span = "P3D"
 
 [manifest]
@@ -25,7 +25,7 @@ constraints = [
     { name = "packaging", specifier = ">=24.0" },
     { name = "setuptools", specifier = ">=83.0.0" },
     { name = "soupsieve", specifier = ">=2.8.4" },
-    { name = "tornado", specifier = ">=6.5.6" },
+    { name = "tornado", specifier = ">=6.5.8" },
 ]
 overrides = [
     { name = "cryptography", specifier = ">=50.0.0,<51.0" },
@@ -4552,7 +4552,7 @@ requires-dist = [
     { name = "pydantic-settings", specifier = ">=2.14.1,<3.0" },
     { name = "pyjwt", marker = "extra == 'proxy'", specifier = ">=2.13.0,<3.0" },
     { name = "pynacl", marker = "extra == 'proxy'", specifier = ">=1.6.2,<2.0" },
-    { name = "pypdf", marker = "extra == 'proxy-runtime'", specifier = ">=6.12.0,<7.0" },
+    { name = "pypdf", marker = "extra == 'proxy-runtime'", specifier = ">=6.16.1,<7.0" },
     { name = "pyroscope-io", marker = "sys_platform != 'win32' and extra == 'proxy'", specifier = ">=0.8.16,<1.0" },
     { name = "python-dotenv", specifier = ">=1.0.0,<2.0" },
     { name = "python-multipart", marker = "extra == 'proxy'", specifier = ">=0.0.27,<1.0" },
@@ -7564,14 +7564,14 @@ wheels = [
 
 [[package]]
 name = "pypdf"
-version = "6.15.0"
+version = "6.16.2"
 source = { registry = "https://pypi.org/simple" }
 dependencies = [
     { name = "typing-extensions", marker = "python_full_version < '3.11'" },
 ]
-sdist = { url = "https://files.pythonhosted.org/packages/17/17/ee75a92718ec7212de831e71454d702225aa5e474a805cce169806044453/pypdf-6.15.0.tar.gz", hash = "sha256:d39c4d955a76409284a905e2d65b40076d77ab76129e0faaeeb6612403ecfc79", size = 6993794, upload-time = "2026-08-06T13:06:49.929Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/44/66/54212e75406afd9f3e933d0dda23072f6aecc55c5a273077dc2e0b028b23/pypdf-6.16.2.tar.gz", hash = "sha256:595647f6191de6f402cfde1d0c455d6cbccbd509aac32b34783009c032de5d6e", size = 7008996, upload-time = "2026-08-23T13:50:07.135Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/af/72/ce3067ac31e214a66388159f8462ddb8c13dd00170f24d555a1f1ae8ee91/pypdf-6.15.0-py3-none-any.whl", hash = "sha256:14e001d6504822cb1ca9c7ed9a69bccb320f59b320730f55af804361abe4d5ee", size = 378123, upload-time = "2026-08-06T13:06:47.709Z" },
+    { url = "https://files.pythonhosted.org/packages/13/f1/a2da3b55acd4ab737bf728c97edaaed5ec1d3c1236acb639dcdfa97e42c7/pypdf-6.16.2-py3-none-any.whl", hash = "sha256:c8b09a59399062fb45a1b8156c18a787a10a3dae03ac9674397a226712c94604", size = 385060, upload-time = "2026-08-23T13:50:05.349Z" },
 ]
 
 [[package]]
@@ -9441,19 +9441,19 @@ wheels = [
 
 [[package]]
 name = "tornado"
-version = "6.5.7"
+version = "6.5.8"
 source = { registry = "https://pypi.org/simple" }
-sdist = { url = "https://files.pythonhosted.org/packages/64/24/95ec527ad67b76d59299e5465b3935d05e4294b7e0290a3924b7487df30b/tornado-6.5.7.tar.gz", hash = "sha256:66c513a76cda70d53907bc27cf1447557699c2e95aa48ba27a442ff61c3ddfc2", size = 519252, upload-time = "2026-06-08T17:34:51.232Z" }
+sdist = { url = "https://files.pythonhosted.org/packages/10/d3/343e5bb989d6515b1646cf3d40135d73f3d5e45339bded401b56cdac24dd/tornado-6.5.8.tar.gz", hash = "sha256:9452e1b208a8bd771e2cb1f2ff564985b9b214bdebbe622793e1799e0a6bd23f", size = 520493, upload-time = "2026-08-07T02:12:42.971Z" }
 wheels = [
-    { url = "https://files.pythonhosted.org/packages/02/dc/c7043cab6fed8ae159fc1923ce829ada35c4dbd797d408a43858ffaf9639/tornado-6.5.7-cp39-abi3-macosx_10_9_universal2.whl", hash = "sha256:148b2eb15c2c765a50796172c1e499649b35f30d2e3c3d3e15913cfa56bfb163", size = 448543, upload-time = "2026-06-08T17:34:38.052Z" },
-    { url = "https://files.pythonhosted.org/packages/92/4f/090b1431e5a43df696feceffc268c5383cc079ecb5f08ce58f917109aafe/tornado-6.5.7-cp39-abi3-macosx_10_9_x86_64.whl", hash = "sha256:9da38de27f1da3b78a966f0dae12b5a1ea9afe72ca805d84ff06508272ddf100", size = 446707, upload-time = "2026-06-08T17:34:39.594Z" },
-    { url = "https://files.pythonhosted.org/packages/37/d8/ef374952fd5da67d4463122c2b8e5a96536ec10b4b339254c6dcde81d01c/tornado-6.5.7-cp39-abi3-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:8d759e71906ee783f8867b93bf26a265743da4c1e2f4a018464c1ba019862972", size = 449774, upload-time = "2026-06-08T17:34:41.204Z" },
-    { url = "https://files.pythonhosted.org/packages/35/37/d434c73f4c6e014b745b9b37085f34f40c022f007efff3d7fe65991899f3/tornado-6.5.7-cp39-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:8a46347a18f23fb92b396beebe0fb78f61dda0cc302445202c16203d8a18848b", size = 450745, upload-time = "2026-06-08T17:34:42.531Z" },
-    { url = "https://files.pythonhosted.org/packages/b6/2b/56b9aff361d7f1ab728a805ec7d7ea835f8807afa9f5cc690ea0e630efb9/tornado-6.5.7-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:7778b30bef919231265e91c69963ce0f49a1e9c07ac900bbe75b19ce2575ba92", size = 450578, upload-time = "2026-06-08T17:34:43.787Z" },
-    { url = "https://files.pythonhosted.org/packages/02/30/a7444fb23aa76860a14198fab96ac79f1866b0a6e19e26c4381b0938e50f/tornado-6.5.7-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:e726f0c75da7726eec023aa62751ff8878bd2737e34fbdd33b1ae5897d2200f5", size = 449985, upload-time = "2026-06-08T17:34:45.326Z" },
-    { url = "https://files.pythonhosted.org/packages/5c/42/5f0e56c01e8d9d36f4e23f367b85ae6cae0c1ecddd5e6977d8388ad27488/tornado-6.5.7-cp39-abi3-win32.whl", hash = "sha256:f8de3bf12d3efdd0cbe7c8887868198f8a91415e3f29fcf258d9b8eb7b1d9ae4", size = 451047, upload-time = "2026-06-08T17:34:46.784Z" },
-    { url = "https://files.pythonhosted.org/packages/c9/a4/b393076ffb21b469eec5b328a0534cf03a3b90bfc6b1f09507cdd075d938/tornado-6.5.7-cp39-abi3-win_amd64.whl", hash = "sha256:de942f843533a039ef9fa3d9c88c7cd8a7c94553fb5ad0154270989b3d99a2c4", size = 451485, upload-time = "2026-06-08T17:34:48.248Z" },
-    { url = "https://files.pythonhosted.org/packages/71/2e/7b1c769803121b809112cf9a00681c472eae1d80e32d7ec0e0bd61d0d0e1/tornado-6.5.7-cp39-abi3-win_arm64.whl", hash = "sha256:ff934fce95643af5f11efdae618eaa73d469dc588641e5c8d19295a0c65c4796", size = 450506, upload-time = "2026-06-08T17:34:49.702Z" },
+    { url = "https://files.pythonhosted.org/packages/f2/d5/007086fd8df5489338e204f65adce33fd4f21a4999dbb2b9cff2f897b5f4/tornado-6.5.8-cp39-abi3-macosx_10_9_universal2.whl", hash = "sha256:cc6aa787d7cfab7c3d35189dc7a56fbd2399a569624c730c6b55b3d6531d0403", size = 449487, upload-time = "2026-08-07T02:12:28.682Z" },
+    { url = "https://files.pythonhosted.org/packages/70/c8/5a24a99495903f594f6a199dd7beead1cbc0a13e2cb9102727bcaaf2a997/tornado-6.5.8-cp39-abi3-macosx_10_9_x86_64.whl", hash = "sha256:9715b5eb79735b2bcd454ce216a9275b7c0470e64ea1bf5742f78b2f72b26eeb", size = 447649, upload-time = "2026-08-07T02:12:30.306Z" },
+    { url = "https://files.pythonhosted.org/packages/6e/de/f2e733f386b85962d1b1dc82cd63d169b5b4580062b35397eac9244a41fe/tornado-6.5.8-cp39-abi3-manylinux1_x86_64.manylinux_2_28_x86_64.manylinux_2_5_x86_64.whl", hash = "sha256:547d63f450d570c14fe0e8db2cfb14c9bbd1c2503b4a6612586267955aa47b58", size = 450707, upload-time = "2026-08-07T02:12:31.95Z" },
+    { url = "https://files.pythonhosted.org/packages/0b/94/20efeee9a01c141e9ac47c397f81679dfda24b32768fc4fff24e76d36c2c/tornado-6.5.8-cp39-abi3-manylinux2014_aarch64.manylinux_2_17_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:7e2360a0ffbe145eca8af0b19cb7203d79b1a98dd4cccdd6b368f6f49c2e3808", size = 451677, upload-time = "2026-08-07T02:12:33.512Z" },
+    { url = "https://files.pythonhosted.org/packages/42/ec/a96ccb8ccf0de2b7bc2c5fa1608a4803735018242e90c4882365a9fd418f/tornado-6.5.8-cp39-abi3-musllinux_1_2_aarch64.whl", hash = "sha256:5d242290bdf7ab3151bc1065fdd75c0dcc21cbc7b49f22a4c56329c2d6566d22", size = 451510, upload-time = "2026-08-07T02:12:35.346Z" },
+    { url = "https://files.pythonhosted.org/packages/29/b5/93185859245ad3f00e62175f29607346788b696369347f0146e0421286bb/tornado-6.5.8-cp39-abi3-musllinux_1_2_x86_64.whl", hash = "sha256:7b94ff0e128fe0542f3bd331fb44d06260fc4ac16881545159f34ef08aad4195", size = 450917, upload-time = "2026-08-07T02:12:36.963Z" },
+    { url = "https://files.pythonhosted.org/packages/97/cf/fe33cf062834487d34d1559746a4a12521033c22645b6d74d4bca702e018/tornado-6.5.8-cp39-abi3-win32.whl", hash = "sha256:67832909c4779c64942380cb5f044a5c6163d00831472d80e25e115de9917836", size = 451952, upload-time = "2026-08-07T02:12:38.512Z" },
+    { url = "https://files.pythonhosted.org/packages/cb/e1/468ad54333e92ccb62627e62cb88e5fc14a2171daa67ed47b1b8542d5b86/tornado-6.5.8-cp39-abi3-win_amd64.whl", hash = "sha256:11881db6b7c168494be2c2d12e65931451bdf7ee718535418ae1d8855dd5a0ee", size = 452391, upload-time = "2026-08-07T02:12:39.971Z" },
+    { url = "https://files.pythonhosted.org/packages/ad/3e/cd5e4f06e34cde33b8ef66cf36aa2b5ad46354cc1af7d2136bbe365fee1d/tornado-6.5.8-cp39-abi3-win_arm64.whl", hash = "sha256:68a7468c7e289f8514d7d664101753903217eff1bb6822c6b5994a0b5f5bcb26", size = 451411, upload-time = "2026-08-07T02:12:41.469Z" },
 ]
 
 [[package]]
diff --git a/whitelisted_bedrock_models.txt b/whitelisted_bedrock_models.txt
index 7e20081988d..8753d7c3c77 100644
--- a/whitelisted_bedrock_models.txt
+++ b/whitelisted_bedrock_models.txt
@@ -217,3 +217,17 @@ bedrock/us-east-1/zai.glm-5
 bedrock/us-west-2/zai.glm-5
 bedrock/us-gov-east-1/anthropic.claude-haiku-4-5-20251001-v1:0
 bedrock/us-gov-west-1/anthropic.claude-haiku-4-5-20251001-v1:0
+bedrock/us-gov-west-1/nvidia.nemotron-nano-3-30b
+bedrock/us-gov-west-1/nvidia.nemotron-nano-12b-v2
+bedrock/us-gov-west-1/nvidia.nemotron-super-3-120b
+bedrock/us-gov-west-1/openai.gpt-oss-20b-1:0
+bedrock/us-gov-west-1/openai.gpt-oss-120b-1:0
+bedrock/us-gov-west-1/anthropic.claude-sonnet-5
+bedrock/us-gov-west-1/anthropic.claude-opus-4-8
+bedrock/us-gov-east-1/nvidia.nemotron-nano-3-30b
+bedrock/us-gov-east-1/nvidia.nemotron-nano-12b-v2
+bedrock/us-gov-east-1/nvidia.nemotron-super-3-120b
+bedrock/us-gov-east-1/openai.gpt-oss-20b-1:0
+bedrock/us-gov-east-1/openai.gpt-oss-120b-1:0
+bedrock/us-gov-east-1/anthropic.claude-sonnet-5
+bedrock/us-gov-east-1/anthropic.claude-opus-4-8