diff --git a/.github/workflows/test-code-quality.yml b/.github/workflows/test-code-quality.yml
index 872a1799d98..1ce77a23810 100644
--- a/.github/workflows/test-code-quality.yml
+++ b/.github/workflows/test-code-quality.yml
@@ -115,6 +115,12 @@ jobs:
       - name: check_fastuuid_usage
         run: uv run --no-sync python ./tests/code_coverage_tests/check_fastuuid_usage.py
 
+      - name: check_e2e_model_freshness_unit_tests
+        run: uv run --no-sync python -m pytest tests/code_coverage_tests/test_check_e2e_model_freshness.py -q
+
+      - name: check_e2e_model_freshness
+        run: uv run --no-sync python ./tests/code_coverage_tests/check_e2e_model_freshness.py
+
       - name: memory_test
         run: uv run --no-sync python ./tests/code_coverage_tests/memory_test.py
 
diff --git a/tests/code_coverage_tests/check_e2e_model_freshness.py b/tests/code_coverage_tests/check_e2e_model_freshness.py
new file mode 100644
index 00000000000..e4c59a89324
--- /dev/null
+++ b/tests/code_coverage_tests/check_e2e_model_freshness.py
@@ -0,0 +1,199 @@
+"""Fail CI when the e2e model matrix (tests/e2e/model_matrix.py) rots.
+
+Checks, in order:
+- every pin resolves to an entry in model_prices_and_context_window.json
+- no pin is within DEPRECATION_WINDOW_DAYS of its deprecation_date
+- the gateway config inside tests/e2e/docker-compose.yml matches GATEWAY_MODELS
+  exactly, and every fallback references a configured alias
+- no test under tests/e2e/ hardcodes a versioned model literal instead of
+  importing a pin from model_matrix.py (triple-quoted strings are exempt as
+  docstrings/prose; the scan is token-based so it survives syntax newer than
+  the interpreter running it)
+"""
+
+import datetime
+import importlib.util
+import io
+import json
+import re
+import sys
+import tokenize
+from pathlib import Path
+from types import ModuleType
+
+import yaml
+from pydantic import BaseModel, ConfigDict
+
+REPO_ROOT = Path(__file__).resolve().parents[2]
+E2E_DIR = REPO_ROOT / "tests" / "e2e"
+MATRIX_PATH = E2E_DIR / "model_matrix.py"
+COMPOSE_PATH = E2E_DIR / "docker-compose.yml"
+PRICING_PATH = REPO_ROOT / "model_prices_and_context_window.json"
+DEPRECATION_WINDOW_DAYS = 30
+MODEL_LITERAL_RE = re.compile(
+    r"(?:gpt|gemini|claude|haiku|sonnet|opus|text-embedding)-\d|rerank-v\d"
+)
+
+
+class PricingEntry(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+    deprecation_date: str | None = None
+
+
+class GatewayLiteLLMParams(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+    model: str
+
+
+class GatewayModelEntry(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+    model_name: str
+    litellm_params: GatewayLiteLLMParams
+
+
+class GatewayRouterSettings(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+    fallbacks: list[dict[str, list[str]]] = []
+
+
+class GatewayConfig(BaseModel):
+    model_config = ConfigDict(extra="ignore")
+    model_list: list[GatewayModelEntry]
+    router_settings: GatewayRouterSettings = GatewayRouterSettings()
+
+
+def load_matrix() -> ModuleType:
+    spec = importlib.util.spec_from_file_location("e2e_model_matrix", MATRIX_PATH)
+    assert spec is not None and spec.loader is not None, f"cannot load {MATRIX_PATH}"
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+def load_pricing() -> dict[str, PricingEntry]:
+    raw = json.loads(PRICING_PATH.read_text(encoding="utf-8"))
+    return {
+        key: PricingEntry.model_validate(value)
+        for key, value in raw.items()
+        if key != "sample_spec"
+    }
+
+
+def load_gateway_config() -> GatewayConfig:
+    compose = yaml.safe_load(COMPOSE_PATH.read_text(encoding="utf-8"))
+    content = compose["configs"]["litellm_config"]["content"]
+    return GatewayConfig.model_validate(yaml.safe_load(content))
+
+
+def resolve_pricing_entry(
+    pricing: dict[str, PricingEntry], canonical: str, model_id: str
+) -> PricingEntry | None:
+    return pricing.get(canonical) or pricing.get(model_id)
+
+
+def existence_violations(matrix: ModuleType, pricing: dict[str, PricingEntry]) -> tuple[str, ...]:
+    return tuple(
+        f"{pin.backend}: neither '{pin.canonical}' nor '{pin.model_id}' is in "
+        f"model_prices_and_context_window.json; the model was removed or the pin is a typo"
+        for pin in matrix.ALL_PINS
+        if resolve_pricing_entry(pricing, pin.canonical, pin.model_id) is None
+    )
+
+
+def deprecation_violations(
+    matrix: ModuleType, pricing: dict[str, PricingEntry], today: datetime.date
+) -> tuple[str, ...]:
+    def violation(pin) -> str | None:
+        entry = resolve_pricing_entry(pricing, pin.canonical, pin.model_id)
+        if entry is None or entry.deprecation_date is None:
+            return None
+        deprecation = datetime.date.fromisoformat(entry.deprecation_date)
+        if (deprecation - today).days > DEPRECATION_WINDOW_DAYS:
+            return None
+        return (
+            f"{pin.backend}: deprecation_date {entry.deprecation_date} is within "
+            f"{DEPRECATION_WINDOW_DAYS} days; bump this pin in tests/e2e/model_matrix.py"
+        )
+
+    return tuple(v for v in (violation(pin) for pin in matrix.ALL_PINS) if v is not None)
+
+
+def gateway_sync_violations(matrix: ModuleType, config: GatewayConfig) -> tuple[str, ...]:
+    expected = frozenset((pin.alias, pin.backend) for pin in matrix.GATEWAY_MODELS)
+    actual = frozenset(
+        (entry.model_name, entry.litellm_params.model) for entry in config.model_list
+    )
+    aliases = frozenset(entry.model_name for entry in config.model_list)
+    missing = tuple(
+        f"docker-compose.yml gateway config is missing model_name '{alias}' -> '{backend}' "
+        f"from GATEWAY_MODELS"
+        for alias, backend in sorted(expected - actual)
+    )
+    extra = tuple(
+        f"docker-compose.yml gateway config has model_name '{alias}' -> '{backend}' "
+        f"that is not in GATEWAY_MODELS; add a pin to tests/e2e/model_matrix.py"
+        for alias, backend in sorted(actual - expected)
+    )
+    dangling_fallbacks = tuple(
+        f"docker-compose.yml fallback references '{name}', which is not a configured model_name"
+        for mapping in config.router_settings.fallbacks
+        for source, targets in mapping.items()
+        for name in (source, *targets)
+        if name not in aliases
+    )
+    return missing + extra + dangling_fallbacks
+
+
+def is_flagged_token(token: tokenize.TokenInfo) -> bool:
+    fstring_middle = getattr(tokenize, "FSTRING_MIDDLE", None)
+    if token.type == fstring_middle:
+        return bool(MODEL_LITERAL_RE.search(token.string))
+    if token.type != tokenize.STRING:
+        return False
+    if token.string.lstrip("rbufRBUF").startswith(('"""', "'''")):
+        return False
+    return bool(MODEL_LITERAL_RE.search(token.string))
+
+
+def literal_violations_in_source(source: str, relative_path: str) -> tuple[str, ...]:
+    tokens = tokenize.generate_tokens(io.StringIO(source).readline)
+    return tuple(
+        f"{relative_path}:{token.start[0]}: hardcoded model literal in {token.string.strip()!r}; "
+        f"import a pin from tests/e2e/model_matrix.py instead"
+        for token in tokens
+        if is_flagged_token(token)
+    )
+
+
+def hardcoded_literal_violations(e2e_dir: Path) -> tuple[str, ...]:
+    return tuple(
+        violation
+        for path in sorted(e2e_dir.rglob("*.py"))
+        if path != MATRIX_PATH
+        for violation in literal_violations_in_source(
+            path.read_text(encoding="utf-8"), str(path.relative_to(REPO_ROOT))
+        )
+    )
+
+
+def main() -> int:
+    matrix = load_matrix()
+    pricing = load_pricing()
+    config = load_gateway_config()
+    violations = (
+        existence_violations(matrix, pricing)
+        + deprecation_violations(matrix, pricing, datetime.date.today())
+        + gateway_sync_violations(matrix, config)
+        + hardcoded_literal_violations(E2E_DIR)
+    )
+    if violations:
+        print("e2e model freshness check failed:")
+        for violation in violations:
+            print(f"  - {violation}")
+        return 1
+    print(f"e2e model freshness check passed for {len(matrix.ALL_PINS)} pins")
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())
diff --git a/tests/code_coverage_tests/test_check_e2e_model_freshness.py b/tests/code_coverage_tests/test_check_e2e_model_freshness.py
new file mode 100644
index 00000000000..7c1f9861edb
--- /dev/null
+++ b/tests/code_coverage_tests/test_check_e2e_model_freshness.py
@@ -0,0 +1,158 @@
+"""Unit tests for check_e2e_model_freshness.py.
+
+Each test feeds the check synthetic inputs and asserts it flags exactly the rot
+it exists to catch: a pin vanishing from the pricing map, a pin nearing its
+deprecation_date, the compose gateway config drifting from GATEWAY_MODELS, and
+a hardcoded model literal sneaking into an e2e test.
+"""
+
+import datetime
+import importlib.util
+from pathlib import Path
+from types import SimpleNamespace
+
+CHECK_PATH = Path(__file__).resolve().parent / "check_e2e_model_freshness.py"
+
+
+def _load_check():
+    spec = importlib.util.spec_from_file_location("check_e2e_model_freshness", CHECK_PATH)
+    assert spec is not None and spec.loader is not None
+    module = importlib.util.module_from_spec(spec)
+    spec.loader.exec_module(module)
+    return module
+
+
+check = _load_check()
+matrix = check.load_matrix()
+ModelPin = matrix.ModelPin
+TODAY = datetime.date(2026, 7, 8)
+
+
+def _matrix_of(*pins, gateway=()):
+    return SimpleNamespace(ALL_PINS=tuple(pins), GATEWAY_MODELS=tuple(gateway))
+
+
+def _pricing(**entries):
+    return {
+        key: check.PricingEntry(deprecation_date=deprecation)
+        for key, deprecation in entries.items()
+    }
+
+
+class TestExistence:
+    def test_pin_missing_from_pricing_map_is_flagged(self):
+        pin = ModelPin("gemini", "gemini-99-flash")
+        violations = check.existence_violations(_matrix_of(pin), _pricing())
+        assert len(violations) == 1
+        assert "gemini/gemini-99-flash" in violations[0]
+
+    def test_pin_resolving_via_prefixed_key_passes(self):
+        pin = ModelPin("gemini", "gemini-99-flash")
+        pricing = _pricing(**{"gemini/gemini-99-flash": None})
+        assert check.existence_violations(_matrix_of(pin), pricing) == ()
+
+    def test_pin_resolving_via_bare_model_id_passes(self):
+        pin = ModelPin("openai", "gpt-99")
+        pricing = _pricing(**{"gpt-99": None})
+        assert check.existence_violations(_matrix_of(pin), pricing) == ()
+
+    def test_pricing_key_override_is_used(self):
+        pin = ModelPin("azure", "my-deployment-name", pricing_key="azure/gpt-99")
+        pricing = _pricing(**{"azure/gpt-99": None})
+        assert check.existence_violations(_matrix_of(pin), pricing) == ()
+
+
+class TestDeprecation:
+    def test_deprecation_within_window_is_flagged(self):
+        pin = ModelPin("openai", "gpt-99")
+        pricing = _pricing(**{"gpt-99": "2026-07-20"})
+        violations = check.deprecation_violations(_matrix_of(pin), pricing, TODAY)
+        assert len(violations) == 1
+        assert "2026-07-20" in violations[0]
+
+    def test_already_deprecated_is_flagged(self):
+        pin = ModelPin("openai", "gpt-99")
+        pricing = _pricing(**{"gpt-99": "2026-01-01"})
+        assert len(check.deprecation_violations(_matrix_of(pin), pricing, TODAY)) == 1
+
+    def test_deprecation_beyond_window_passes(self):
+        pin = ModelPin("openai", "gpt-99")
+        pricing = _pricing(**{"gpt-99": "2027-01-01"})
+        assert check.deprecation_violations(_matrix_of(pin), pricing, TODAY) == ()
+
+    def test_no_deprecation_date_passes(self):
+        pin = ModelPin("openai", "gpt-99")
+        pricing = _pricing(**{"gpt-99": None})
+        assert check.deprecation_violations(_matrix_of(pin), pricing, TODAY) == ()
+
+
+class TestGatewaySync:
+    def _config(self, models, fallbacks=()):
+        return check.GatewayConfig(
+            model_list=[
+                check.GatewayModelEntry(
+                    model_name=name,
+                    litellm_params=check.GatewayLiteLLMParams(model=backend),
+                )
+                for name, backend in models
+            ],
+            router_settings=check.GatewayRouterSettings(fallbacks=list(fallbacks)),
+        )
+
+    def test_matching_config_passes(self):
+        pin = ModelPin("gemini", "gemini-99-flash")
+        config = self._config([(pin.alias, pin.backend)])
+        assert check.gateway_sync_violations(_matrix_of(gateway=(pin,)), config) == ()
+
+    def test_stale_compose_model_is_flagged_both_ways(self):
+        pin = ModelPin("gemini", "gemini-99-flash")
+        config = self._config([("gemini-98-flash", "gemini/gemini-98-flash")])
+        violations = check.gateway_sync_violations(_matrix_of(gateway=(pin,)), config)
+        assert any("missing model_name 'gemini-99-flash'" in v for v in violations)
+        assert any("has model_name 'gemini-98-flash'" in v for v in violations)
+
+    def test_fallback_referencing_unknown_alias_is_flagged(self):
+        pin = ModelPin("gemini", "gemini-99-flash")
+        config = self._config(
+            [(pin.alias, pin.backend)],
+            fallbacks=[{pin.alias: ["gpt-99"]}],
+        )
+        violations = check.gateway_sync_violations(_matrix_of(gateway=(pin,)), config)
+        assert len(violations) == 1
+        assert "gpt-99" in violations[0]
+
+
+class TestLiteralScan:
+    def test_hardcoded_model_literal_is_flagged(self):
+        source = 'MODEL = "gemini-2.5-flash"\n'
+        violations = check.literal_violations_in_source(source, "suite/test_x.py")
+        assert len(violations) == 1
+        assert "suite/test_x.py:1" in violations[0]
+
+    def test_pin_usage_is_not_flagged(self):
+        source = (
+            "from model_matrix import GEMINI_CHAT\n"
+            "MODEL = GEMINI_CHAT.alias\n"
+            'MESSAGE = f"model {GEMINI_CHAT.alias} denied"\n'
+        )
+        assert check.literal_violations_in_source(source, "suite/test_x.py") == ()
+
+    def test_docstrings_and_comments_are_exempt(self):
+        source = (
+            '"""Drives gemini-2.5-flash against the proxy."""\n'
+            "MODEL = None  # was gemini-2.5-flash\n"
+        )
+        assert check.literal_violations_in_source(source, "suite/test_x.py") == ()
+
+    def test_rerank_literal_is_flagged(self):
+        source = 'MODEL = "cohere/rerank-v3.5"\n'
+        assert len(check.literal_violations_in_source(source, "suite/test_x.py")) == 1
+
+    def test_syntax_newer_than_interpreter_is_still_scanned(self):
+        source = (
+            "def unwrap[R](result: Result[R]) -> R: ...\n"
+            'MODEL = "gpt-4o-mini"\n'
+        )
+        violations = check.literal_violations_in_source(source, "suite/test_x.py")
+        assert len(violations) == 1
+        assert "suite/test_x.py:2" in violations[0]
diff --git a/tests/e2e/CLAUDE.md b/tests/e2e/CLAUDE.md
index a4c507ca5ea..9d6cb72834b 100644
--- a/tests/e2e/CLAUDE.md
+++ b/tests/e2e/CLAUDE.md
@@ -32,13 +32,17 @@ class TestPromptCompression:
 
     def test_prompt_compression_accumulate_spend(self, key_id, user_id):
         for _ in range(10):
-            response = self.resources.gateway.post("gemini-2.5-flash", key_id, user_id)
+            response = self.resources.gateway.post(GEMINI_CHAT.alias, key_id, user_id)
         compressed_value = ...
         assert response.cost == compressed_value  # the cost was actually reduced
 ```
 
 That snippet only conveys intent. What you actually write uses the real harness: the `client` fixture for your suite, the `scoped_key` fixture for an auto-deleted key, typed pydantic bodies from `models.py`, and `unwrap(...)` on the tagged-union result. `tests/e2e/llm_translation/test_custom_pricing_e2e.py` is the reference to copy from; it creates a scoped key, drives a real gemini call, polls `/spend/logs` to a deadline for the cost-breakdown row, then asserts the input and output costs match the configured custom rates and that a sibling deployment kept its own price. Read it before writing yours
 
+## Models come from model_matrix.py, never hardcoded
+
+Every model a test drives is a pin imported from `model_matrix.py` (`GEMINI_CHAT.alias`, `OPENAI_CHAT_MINI.backend`, ...), named for the role it plays, never its version. Bumping a model version is then a one-file change (plus `docker-compose.yml`, which cannot import Python). `tests/code_coverage_tests/check_e2e_model_freshness.py` enforces this in CI: it fails on hardcoded versioned model literals in any `.py` file here, on pins missing from `model_prices_and_context_window.json` or near their `deprecation_date`, and on drift between `GATEWAY_MODELS` and the compose gateway config
+
 ## Use the shared transport; never touch requests directly
 
 Every HTTP call goes through the shared transport, never through `requests.*` in a test. `e2e_http.py` is the only module permitted to call `requests.*`, and that is enforced in CI by `tests/code_coverage_tests/check_e2e_no_raw_requests.py`. A test that imports requests will fail the check
diff --git a/tests/e2e/CONTRIBUTING.md b/tests/e2e/CONTRIBUTING.md
index 4395d299c5b..784a8b4b4b3 100644
--- a/tests/e2e/CONTRIBUTING.md
+++ b/tests/e2e/CONTRIBUTING.md
@@ -9,7 +9,7 @@ When contributing to this directory, please first discuss the change you wish to
 
 ## Setup
 
-The suites run against a live proxy, so bring one up first. `docker-compose.yml` here starts that proxy with a throwaway Postgres and Redis; `docker compose down -v` resets everything, so no state leaks between runs. The proxy config is inlined in the compose file under `configs`, prewired with example models (`gpt-5.5`, `claude-haiku-4-5`, `gemini-2.5-flash`, `openai-text-embedding-3-small`) whose keys come from your `.env`. If your test needs another model, a pricing override, or a guardrail declared up front, add it to that inline config and read it back in the test rather than hardcoding values
+The suites run against a live proxy, so bring one up first. `docker-compose.yml` here starts that proxy with a throwaway Postgres and Redis; `docker compose down -v` resets everything, so no state leaks between runs. The proxy config is inlined in the compose file under `configs`, prewired with the models pinned in `model_matrix.py` (kept in sync by `tests/code_coverage_tests/check_e2e_model_freshness.py`) whose keys come from your `.env`. If your test needs another model, a pricing override, or a guardrail declared up front, add it to that inline config and read it back in the test rather than hardcoding values
 
 ## Running the tests locally
 
@@ -113,7 +113,7 @@ class TestPromptCompression:
 
     def test_prompt_compression_accumulate_spend(self, key_id, user_id):
         for _ in range(10):
-            response = self.resources.gateway.post("gemini-2.5-flash", key_id, user_id)
+            response = self.resources.gateway.post(GEMINI_CHAT.alias, key_id, user_id)
         compressed_value = ...
         assert response.cost == compressed_value  # the cost was actually reduced
 ```
diff --git a/tests/e2e/access_control/access_control_client.py b/tests/e2e/access_control/access_control_client.py
index d7bc9c280aa..7f2ead9ff79 100644
--- a/tests/e2e/access_control/access_control_client.py
+++ b/tests/e2e/access_control/access_control_client.py
@@ -6,6 +6,7 @@ from dataclasses import dataclass
 
 from e2e_gateway import Gateway, build_gateway
 from e2e_http import StreamingResponse
+from model_matrix import OPENAI_CHAT_MINI
 from models import (
     ChatBody,
     ChatMessage,
@@ -46,7 +47,7 @@ class AccessControlClient:
             headers=self.gateway.transport.bearer(key),
             json=ModelNewBody(
                 model_name=model_name,
-                litellm_params=LiteLLMParamsBody(model="openai/gpt-4o-mini"),
+                litellm_params=LiteLLMParamsBody(model=OPENAI_CHAT_MINI.backend),
                 model_info=ModelInfoBody(id=model_name),
             ),
         )
diff --git a/tests/e2e/access_control/test_access_control_e2e.py b/tests/e2e/access_control/test_access_control_e2e.py
index ce649fa2400..68846493e21 100644
--- a/tests/e2e/access_control/test_access_control_e2e.py
+++ b/tests/e2e/access_control/test_access_control_e2e.py
@@ -24,11 +24,12 @@ from access_control_client import (
 )
 from e2e_config import unique_marker
 from lifecycle import ResourceManager
+from model_matrix import GEMINI_CHAT, OPENAI_CHAT
 
 pytestmark = pytest.mark.e2e
 
-ALLOWED_MODEL = "gemini-2.5-flash"
-DISALLOWED_MODEL = "gpt-5.5"
+ALLOWED_MODEL = GEMINI_CHAT.alias
+DISALLOWED_MODEL = OPENAI_CHAT.alias
 
 
 def _is_json(body: str) -> bool:
diff --git a/tests/e2e/batches/capabilities.py b/tests/e2e/batches/capabilities.py
index 522e3162e24..2d45b5cc8bf 100644
--- a/tests/e2e/batches/capabilities.py
+++ b/tests/e2e/batches/capabilities.py
@@ -13,6 +13,7 @@ import base64
 from dataclasses import dataclass
 from typing import Literal
 
+from model_matrix import AZURE_BATCH, BEDROCK_ANTHROPIC_CHAT, GEMINI_CHAT, OPENAI_CHAT_MINI, VERTEX_CHAT
 from models import LiteLLMParamsBody
 
 Scenario = Literal["encoded", "unified", "model_param", "provider_fallback"]
@@ -39,26 +40,26 @@ class Provider:
         match self.name:
             case "openai":
                 return LiteLLMParamsBody(
-                    model="openai/gpt-4o-mini",
+                    model=OPENAI_CHAT_MINI.backend,
                     api_key="os.environ/OPENAI_API_KEY",
                 )
             case "azure":
                 return LiteLLMParamsBody(
-                    model="azure/gpt-4.1-mini-batch",
+                    model=AZURE_BATCH.backend,
                     api_base="os.environ/AZURE_API_BASE",
                     api_key="os.environ/AZURE_API_KEY",
                     api_version="2024-07-01-preview",
                 )
             case "vertex_ai":
                 return LiteLLMParamsBody(
-                    model="vertex_ai/gemini-2.5-flash",
+                    model=VERTEX_CHAT.backend,
                     vertex_project="os.environ/VERTEXAI_PROJECT",
                     vertex_location="us-central1",
                     vertex_credentials="os.environ/VERTEXAI_CREDENTIALS",
                 )
             case "bedrock":
                 return LiteLLMParamsBody(
-                    model="bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
+                    model=BEDROCK_ANTHROPIC_CHAT.backend,
                     s3_access_key_id="os.environ/AWS_ACCESS_KEY_ID",
                     s3_secret_access_key="os.environ/AWS_SECRET_ACCESS_KEY",
                     s3_region_name="os.environ/AWS_REGION",
@@ -96,15 +97,15 @@ class Capability:
 
 
 PROVIDERS: tuple[Provider, ...] = (
-    Provider("openai", "openai-batch", "gpt-4o-mini", can_cancel=True, can_list=True),
-    Provider("azure", "azure-batch", "gpt-4.1-mini-batch", can_cancel=True, can_list=True),
+    Provider("openai", "openai-batch", OPENAI_CHAT_MINI.alias, can_cancel=True, can_list=True),
+    Provider("azure", "azure-batch", AZURE_BATCH.alias, can_cancel=True, can_list=True),
     Provider(
-        "vertex_ai", "vertex-batch", "gemini-2.5-flash", can_cancel=True, can_list=True
+        "vertex_ai", "vertex-batch", GEMINI_CHAT.alias, can_cancel=True, can_list=True
     ),
     Provider(
         "bedrock",
         "bedrock-batch",
-        "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
+        BEDROCK_ANTHROPIC_CHAT.backend,
         can_cancel=False,
         can_list=False,
     ),
diff --git a/tests/e2e/batches/test_batches_e2e.py b/tests/e2e/batches/test_batches_e2e.py
index a998f962c04..a89fe13288b 100644
--- a/tests/e2e/batches/test_batches_e2e.py
+++ b/tests/e2e/batches/test_batches_e2e.py
@@ -49,6 +49,7 @@ from e2e_http import (
     unwrap,
 )
 from lifecycle import ResourceManager
+from model_matrix import OPENAI_CHAT_MINI
 from models import KeyGenerateBody, SpendLogRow, SpendLogsParams
 
 pytestmark = pytest.mark.e2e
@@ -339,7 +340,7 @@ def test_rate_limited_batch_create_leaves_no_unattributed_spend_row(
 
     file = unwrap(
         client.upload_file(
-            content=render_jsonl("gpt-4o-mini"),
+            content=render_jsonl(OPENAI_CHAT_MINI.alias),
             form=FileUploadForm(purpose="batch"),
             model="openai-batch",
             key=key,
diff --git a/tests/e2e/budgets/test_budget_enforcement_e2e.py b/tests/e2e/budgets/test_budget_enforcement_e2e.py
index d03288b637a..6f40191312a 100644
--- a/tests/e2e/budgets/test_budget_enforcement_e2e.py
+++ b/tests/e2e/budgets/test_budget_enforcement_e2e.py
@@ -20,6 +20,7 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import run_case
+from model_matrix import ANTHROPIC_CHAT
 
 pytestmark = pytest.mark.e2e
 
@@ -31,7 +32,7 @@ def _assert_budget_blocks(client: BudgetClient, key: str, *, user: str = "") ->
     for _ in range(40):
         result = client.chat(
             key,
-            "claude-haiku-4-5",
+            ANTHROPIC_CHAT.alias,
             f"spend {unique_marker()}",
             max_tokens=16,
             user=user or None,
@@ -88,7 +89,7 @@ class EndUserBudgetCase(_BudgetCase):
         customer = f"e2e-budget-cust-{unique_marker()}"
         self.client.create_customer(customer, max_budget=3e-6)
         self._undo.append(lambda: self.client.delete_customers([customer]))
-        self.key = self.client.generate_key(models=["claude-haiku-4-5"])
+        self.key = self.client.generate_key(models=[ANTHROPIC_CHAT.alias])
         self._undo.append(lambda: self.client.delete_key(self.key))
         self._customer = customer
 
diff --git a/tests/e2e/budgets/test_budget_fallback_e2e.py b/tests/e2e/budgets/test_budget_fallback_e2e.py
index 6114bfe6fc9..c8a993c8e44 100644
--- a/tests/e2e/budgets/test_budget_fallback_e2e.py
+++ b/tests/e2e/budgets/test_budget_fallback_e2e.py
@@ -12,11 +12,12 @@ import pytest
 from budget_client import BudgetClient, model_budget
 from e2e_config import unique_marker
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT, OPENAI_CHAT
 
 pytestmark = pytest.mark.e2e
 
-PRIMARY_MODEL = "claude-haiku-4-5"
-FALLBACK_MODEL = "gpt-5.5"
+PRIMARY_MODEL = ANTHROPIC_CHAT.alias
+FALLBACK_MODEL = OPENAI_CHAT.alias
 
 
 def test_budget_fallback_reroutes_anthropic_messages_to_openai(
diff --git a/tests/e2e/budgets/test_budget_reset_e2e.py b/tests/e2e/budgets/test_budget_reset_e2e.py
index dcf776db9a2..d970a0a7199 100644
--- a/tests/e2e/budgets/test_budget_reset_e2e.py
+++ b/tests/e2e/budgets/test_budget_reset_e2e.py
@@ -16,13 +16,14 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 
 pytestmark = pytest.mark.e2e
 
 
 def _call(client: BudgetClient, key: str):
     return client.chat(
-        key, "claude-haiku-4-5", f"reset {unique_marker()}", max_tokens=16
+        key, ANTHROPIC_CHAT.alias, f"reset {unique_marker()}", max_tokens=16
     )
 
 
diff --git a/tests/e2e/budgets/test_model_max_budget_e2e.py b/tests/e2e/budgets/test_model_max_budget_e2e.py
index 44e6a333ef0..6e2a09e371f 100644
--- a/tests/e2e/budgets/test_model_max_budget_e2e.py
+++ b/tests/e2e/budgets/test_model_max_budget_e2e.py
@@ -14,11 +14,12 @@ from budget_client import BudgetClient, is_budget_block, model_budget
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT, GEMINI_CHAT
 
 pytestmark = pytest.mark.e2e
 
-CAPPED_MODEL = "claude-haiku-4-5"
-FREE_MODEL = "gemini-2.5-flash"
+CAPPED_MODEL = ANTHROPIC_CHAT.alias
+FREE_MODEL = GEMINI_CHAT.alias
 
 
 def _call(client: BudgetClient, key: str, model: str):
diff --git a/tests/e2e/budgets/test_multi_window_budget_e2e.py b/tests/e2e/budgets/test_multi_window_budget_e2e.py
index 553ad1ce701..6ea894afb69 100644
--- a/tests/e2e/budgets/test_multi_window_budget_e2e.py
+++ b/tests/e2e/budgets/test_multi_window_budget_e2e.py
@@ -16,6 +16,7 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 from models import BudgetWindow
 
 pytestmark = pytest.mark.e2e
@@ -25,7 +26,7 @@ WINDOW_SECONDS = 30  # the tight window; calls succeed again only after it elaps
 
 def _call(client: BudgetClient, key: str):
     return client.chat(
-        key, "claude-haiku-4-5", f"window {unique_marker()}", max_tokens=16
+        key, ANTHROPIC_CHAT.alias, f"window {unique_marker()}", max_tokens=16
     )
 
 
diff --git a/tests/e2e/budgets/test_soft_budget_e2e.py b/tests/e2e/budgets/test_soft_budget_e2e.py
index 407de7ae467..da4212e5058 100644
--- a/tests/e2e/budgets/test_soft_budget_e2e.py
+++ b/tests/e2e/budgets/test_soft_budget_e2e.py
@@ -13,6 +13,7 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 
 pytestmark = pytest.mark.e2e
 
@@ -26,7 +27,7 @@ def test_soft_budget_does_not_block(
 
     for _ in range(3):
         result = client.chat(
-            key, "claude-haiku-4-5", f"hi {unique_marker()}", max_tokens=16
+            key, ANTHROPIC_CHAT.alias, f"hi {unique_marker()}", max_tokens=16
         )
         assert not is_budget_block(result), (
             "soft_budget blocked a request; it must alert only, not block "
diff --git a/tests/e2e/budgets/test_spend_counter_reseed_e2e.py b/tests/e2e/budgets/test_spend_counter_reseed_e2e.py
index a6860aeef43..c8b9987babf 100644
--- a/tests/e2e/budgets/test_spend_counter_reseed_e2e.py
+++ b/tests/e2e/budgets/test_spend_counter_reseed_e2e.py
@@ -28,6 +28,7 @@ from budget_client import BudgetClient
 from e2e_config import unique_marker
 from e2e_http import StreamingResponse
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 
 if TYPE_CHECKING:
     import redis
@@ -35,7 +36,7 @@ if TYPE_CHECKING:
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "claude-haiku-4-5"
+MODEL = ANTHROPIC_CHAT.alias
 ACCUMULATE_CALLS = 24
 BURST = 6
 # proxy_batch_write_at (60s) flushes the spend to the DB and default_redis_ttl (20s)
diff --git a/tests/e2e/budgets/test_tag_budget_e2e.py b/tests/e2e/budgets/test_tag_budget_e2e.py
index 7cec5bc96c1..5536308a450 100644
--- a/tests/e2e/budgets/test_tag_budget_e2e.py
+++ b/tests/e2e/budgets/test_tag_budget_e2e.py
@@ -14,6 +14,7 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 
 pytestmark = pytest.mark.e2e
 
@@ -23,7 +24,7 @@ TINY_BUDGET = 1e-6
 def _tagged_call(client: BudgetClient, key: str, tag: str):
     result = client.chat(
         key,
-        "claude-haiku-4-5",
+        ANTHROPIC_CHAT.alias,
         f"hi {unique_marker()}",
         tags=[tag],
         max_tokens=16,
diff --git a/tests/e2e/budgets/test_team_member_budget_e2e.py b/tests/e2e/budgets/test_team_member_budget_e2e.py
index 301617bfdca..680966dede3 100644
--- a/tests/e2e/budgets/test_team_member_budget_e2e.py
+++ b/tests/e2e/budgets/test_team_member_budget_e2e.py
@@ -22,11 +22,12 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import Success, require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 from models import ChatBody, ChatMessage
 
 pytestmark = pytest.mark.e2e
 
-MODEL = "claude-haiku-4-5"
+MODEL = ANTHROPIC_CHAT.alias
 TEAM_BUDGET = 100.0
 MEMBER_BUDGET = 3e-6
 BURST = 6
diff --git a/tests/e2e/budgets/test_team_member_budget_reset_e2e.py b/tests/e2e/budgets/test_team_member_budget_reset_e2e.py
index 2749f16a26e..d09a7818d62 100644
--- a/tests/e2e/budgets/test_team_member_budget_reset_e2e.py
+++ b/tests/e2e/budgets/test_team_member_budget_reset_e2e.py
@@ -7,6 +7,7 @@ from budget_client import BudgetClient
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 
 pytestmark = pytest.mark.e2e
 
@@ -33,7 +34,7 @@ def test_team_member_budget_reset_keeps_advancing(client: BudgetClient, resource
     # the member can spend within the team while the window is live
     key = client.generate_key(team_id=team_id, user_id=user_id)
     resources.defer(lambda: client.delete_key(key))
-    require_successful_call(client.chat(key, "claude-haiku-4-5", f"reset {unique_marker()}", max_tokens=16))
+    require_successful_call(client.chat(key, ANTHROPIC_CHAT.alias, f"reset {unique_marker()}", max_tokens=16))
 
     # once the window elapses the reset job must move budget_reset_at forward; a job
     # that skips the member's budget row (the #25109 regression) leaves it pinned at
diff --git a/tests/e2e/budgets/test_team_multi_window_budget_e2e.py b/tests/e2e/budgets/test_team_multi_window_budget_e2e.py
index c58e74db965..6c7d62223b3 100644
--- a/tests/e2e/budgets/test_team_multi_window_budget_e2e.py
+++ b/tests/e2e/budgets/test_team_multi_window_budget_e2e.py
@@ -21,6 +21,7 @@ from budget_client import BudgetClient, is_budget_block
 from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 from models import BudgetWindow
 
 pytestmark = pytest.mark.e2e
@@ -29,7 +30,7 @@ WINDOW_SECONDS = 30
 
 
 def _call(client: BudgetClient, key: str):
-    return client.chat(key, "claude-haiku-4-5", f"team-window {unique_marker()}", max_tokens=16)
+    return client.chat(key, ANTHROPIC_CHAT.alias, f"team-window {unique_marker()}", max_tokens=16)
 
 
 def test_team_short_window_blocks_then_resets(client: BudgetClient, resources: ResourceManager) -> None:
diff --git a/tests/e2e/docker-compose.yml b/tests/e2e/docker-compose.yml
index cdf5d6cbf6f..9b4e758c7fb 100644
--- a/tests/e2e/docker-compose.yml
+++ b/tests/e2e/docker-compose.yml
@@ -23,7 +23,7 @@ configs:
         allowed_fails: 5
         cooldown_time: 30
         fallbacks:
-          - gemini-2.5-flash: ["gpt-5.5", "claude-haiku-4-5"]
+          - gemini-3.5-flash: ["gpt-5.5", "claude-haiku-4-5"]
 
       model_list:
         - model_name: gpt-5.5
@@ -36,9 +36,9 @@ configs:
             model: anthropic/claude-haiku-4-5
             api_key: os.environ/ANTHROPIC_API_KEY
 
-        - model_name: gemini-2.5-flash
+        - model_name: gemini-3.5-flash
           litellm_params:
-            model: gemini/gemini-2.5-flash
+            model: gemini/gemini-3.5-flash
             api_key: os.environ/GEMINI_API_KEY
 
         - model_name: openai-text-embedding-3-small
diff --git a/tests/e2e/llm_translation/test_audio_speech_e2e.py b/tests/e2e/llm_translation/test_audio_speech_e2e.py
index f7a04d94cb3..8187d1c368b 100644
--- a/tests/e2e/llm_translation/test_audio_speech_e2e.py
+++ b/tests/e2e/llm_translation/test_audio_speech_e2e.py
@@ -13,6 +13,7 @@ from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from endpoints_client import EndpointsClient
 from lifecycle import ResourceManager
+from model_matrix import OPENAI_TTS
 from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
@@ -26,7 +27,7 @@ class TestAudioSpeech:
         model_id = endpoints_client.create_model(
             model,
             LiteLLMParamsBody(
-                model="openai/gpt-4o-mini-tts", api_key="os.environ/OPENAI_API_KEY"
+                model=OPENAI_TTS.backend, api_key="os.environ/OPENAI_API_KEY"
             ),
         )
         resources.defer(lambda: endpoints_client.delete_model(model_id))
diff --git a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
index 5cc4ff308fa..efa57ccd2c5 100644
--- a/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
+++ b/tests/e2e/llm_translation/test_chat_completions_regression_e2e.py
@@ -15,15 +15,16 @@ import pytest
 
 from e2e_config import unique_marker
 from e2e_http import unwrap
+from model_matrix import ANTHROPIC_CHAT, GEMINI_CHAT, OPENAI_CHAT
 from models import ChatBody, ChatMessage
 from passthrough_client import PassthroughClient
 
 pytestmark = pytest.mark.e2e
 
 CHAT_MODELS: tuple[tuple[str, str], ...] = (
-    ("gpt-5.5", "openai"),
-    ("claude-haiku-4-5", "anthropic"),
-    ("gemini-2.5-flash", "gemini"),
+    (OPENAI_CHAT.alias, "openai"),
+    (ANTHROPIC_CHAT.alias, "anthropic"),
+    (GEMINI_CHAT.alias, "gemini"),
 )
 
 
diff --git a/tests/e2e/llm_translation/test_custom_pricing_e2e.py b/tests/e2e/llm_translation/test_custom_pricing_e2e.py
index 7894b447be9..15b55bcb4f4 100644
--- a/tests/e2e/llm_translation/test_custom_pricing_e2e.py
+++ b/tests/e2e/llm_translation/test_custom_pricing_e2e.py
@@ -5,12 +5,13 @@ teardown) instead of relying on a statically configured model, so the check is
 self-contained and never inherits pricing another suite or a stale config left on
 the shared proxy. custom-priced-flash sets input/output rates deliberately far
 above the canonical gemini price; the isolation sibling shares the same
-gemini/gemini-2.5-flash backend but sets no override. Three behaviors are checked
-independently:
+gemini flash backend (GEMINI_CHAT in model_matrix.py) but sets no override.
+Three behaviors are checked independently:
 
 - billing: a real call's logged cost breakdown charges input and output tokens at
-  the custom rates, each component checked separately (a base-rate bill lands
-  ~100x lower; a swapped input/output rate passes a total-only check but not this)
+  the custom rates, each component checked separately (a base-rate bill lands an
+  order of magnitude lower; a swapped input/output rate passes a total-only check
+  but not this)
 - reporting: /model/info surfaces those rates for the deployment
 - isolation: the sibling keeps its own price; an override that leaks into the
   shared backend cost map (LIT-3897) misprices it, making the sibling's rate match
@@ -27,6 +28,7 @@ from e2e_gateway import Gateway
 from e2e_http import Success, unwrap
 from endpoints_client import EndpointsClient
 from lifecycle import ResourceManager
+from model_matrix import GEMINI_CHAT
 from models import (
     ChatBody,
     ChatMessage,
@@ -37,9 +39,9 @@ from models import (
 
 pytestmark = pytest.mark.e2e
 
-BACKEND_MODEL = "gemini/gemini-2.5-flash"
+BACKEND_MODEL = GEMINI_CHAT.backend
 GEMINI_API_KEY = "os.environ/GEMINI_API_KEY"
-# Deliberately ~100x above canonical gemini-2.5-flash (input 3e-7 / output 2.5e-6)
+# Deliberately an order of magnitude above the canonical gemini flash rates
 # so an override that is ignored or under-applied bills at the base rate and fails.
 CUSTOM_INPUT_RATE = 5e-05
 CUSTOM_OUTPUT_RATE = 1e-04
@@ -78,7 +80,7 @@ def _provision(
     input_cost_per_token: float | None,
     output_cost_per_token: float | None,
 ) -> str:
-    """Register a fresh gemini/gemini-2.5-flash deployment (deleted on teardown) and
+    """Register a fresh gemini flash deployment (deleted on teardown) and
     return its model name. With the cost fields set the deployment carries a custom
     pricing override; with them None it is a plain sibling on the same backend. The
     marker keeps the name unique so concurrent runs on the shared proxy never
@@ -230,7 +232,7 @@ class TestCustomPricing:
         assert sibling_entry is not None, f"{sibling} absent from /model/info"
 
         # custom-priced-flash overrides pricing; the sibling shares the same
-        # gemini/gemini-2.5-flash backend but sets no override, so it must keep its
+        # gemini flash backend but sets no override, so it must keep its
         # own price. Equal rates mean the override leaked into the shared cost map.
         assert (
             sibling_entry.model_info.input_cost_per_token
diff --git a/tests/e2e/llm_translation/test_embeddings_endpoint_e2e.py b/tests/e2e/llm_translation/test_embeddings_endpoint_e2e.py
index 56f2de8bd4f..4e990868a98 100644
--- a/tests/e2e/llm_translation/test_embeddings_endpoint_e2e.py
+++ b/tests/e2e/llm_translation/test_embeddings_endpoint_e2e.py
@@ -14,6 +14,7 @@ from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from endpoints_client import EmbeddingsResult, EndpointsClient
 from lifecycle import ResourceManager
+from model_matrix import OPENAI_EMBEDDING
 from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
@@ -27,7 +28,7 @@ class TestEmbeddingsEndpoint:
         model_id = endpoints_client.create_model(
             model,
             LiteLLMParamsBody(
-                model="openai/text-embedding-3-small", api_key="os.environ/OPENAI_API_KEY"
+                model=OPENAI_EMBEDDING.backend, api_key="os.environ/OPENAI_API_KEY"
             ),
         )
         resources.defer(lambda: endpoints_client.delete_model(model_id))
diff --git a/tests/e2e/llm_translation/test_messages_e2e.py b/tests/e2e/llm_translation/test_messages_e2e.py
index b0a48f22118..65999f76a32 100644
--- a/tests/e2e/llm_translation/test_messages_e2e.py
+++ b/tests/e2e/llm_translation/test_messages_e2e.py
@@ -13,6 +13,7 @@ from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from endpoints_client import EndpointsClient, MessagesResult
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT
 from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
@@ -26,7 +27,7 @@ class TestAnthropicMessages:
         model_id = endpoints_client.create_model(
             model,
             LiteLLMParamsBody(
-                model="anthropic/claude-haiku-4-5", api_key="os.environ/ANTHROPIC_API_KEY"
+                model=ANTHROPIC_CHAT.backend, api_key="os.environ/ANTHROPIC_API_KEY"
             ),
         )
         resources.defer(lambda: endpoints_client.delete_model(model_id))
diff --git a/tests/e2e/llm_translation/test_passthrough_e2e.py b/tests/e2e/llm_translation/test_passthrough_e2e.py
index 37d55c665b3..a4787d6903f 100644
--- a/tests/e2e/llm_translation/test_passthrough_e2e.py
+++ b/tests/e2e/llm_translation/test_passthrough_e2e.py
@@ -4,7 +4,7 @@ Each test sends a NATIVE provider request through the proxy's passthrough route
 and verifies the proxy still logged a costed SpendLogs row
 (call_type="pass_through_endpoint"), correlated by the x-litellm-call-id header.
 
-Covered: gemini ("gemini-2.5-flash") + anthropic ("claude-haiku-4-5"), streaming +
+Covered: gemini (GEMINI_CHAT.alias) + anthropic (ANTHROPIC_CHAT.alias), streaming +
 non-streaming, plus native tool calls. See LLM_TRANSLATION_COVERAGE_MATRIX.md.
 
 A passthrough call returning non-2xx fails hard (never a skip); once it returns
@@ -15,6 +15,7 @@ import pytest
 
 from e2e_config import unique_marker
 from e2e_http import StreamingResponse, require_successful_call
+from model_matrix import ANTHROPIC_CHAT, GEMINI_CHAT
 from models import SpendLogRow
 from passthrough_client import (
     AnthropicTool,
@@ -55,7 +56,7 @@ def test_gemini_passthrough_nonstreaming_logs_cost(
 ) -> None:
     tag = f"e2e-passthrough-{unique_marker()}"
     result = client.gemini_generate(
-        scoped_key, "gemini-2.5-flash", "Say hello in one word", tags=[tag, "gemini"]
+        scoped_key, GEMINI_CHAT.alias, "Say hello in one word", tags=[tag, "gemini"]
     )
     require_successful_call(result)
 
@@ -68,7 +69,7 @@ def test_gemini_passthrough_nonstreaming_logs_cost(
 def test_gemini_passthrough_streaming_logs_cost(
     client: PassthroughClient, scoped_key: str
 ) -> None:
-    result = client.gemini_stream(scoped_key, "gemini-2.5-flash", "Count to five")
+    result = client.gemini_stream(scoped_key, GEMINI_CHAT.alias, "Count to five")
     require_successful_call(result)
     assert result.chunks > 0, "streaming passthrough produced no events"
 
@@ -81,7 +82,7 @@ def test_gemini_passthrough_tool_call_logs_cost(
 ) -> None:
     result = client.gemini_generate(
         scoped_key,
-        "gemini-2.5-flash",
+        GEMINI_CHAT.alias,
         "What is the weather in Paris? Use the get_weather tool.",
         tools=[
             GeminiTool(
@@ -112,7 +113,7 @@ def test_gemini_passthrough_tool_call_logs_cost(
 def test_anthropic_passthrough_nonstreaming_logs_cost(
     client: PassthroughClient, scoped_key: str
 ) -> None:
-    result = client.anthropic_message(scoped_key, "claude-haiku-4-5", "Say hello")
+    result = client.anthropic_message(scoped_key, ANTHROPIC_CHAT.alias, "Say hello")
     require_successful_call(result)
 
     row = _fetch_cost_breakdown(client, result)
@@ -124,7 +125,7 @@ def test_anthropic_passthrough_streaming_logs_cost(
     client: PassthroughClient, scoped_key: str
 ) -> None:
     result = client.anthropic_message(
-        scoped_key, "claude-haiku-4-5", "Count to five", stream=True
+        scoped_key, ANTHROPIC_CHAT.alias, "Count to five", stream=True
     )
     require_successful_call(result)
     assert result.chunks > 0, "streaming passthrough produced no events"
@@ -138,7 +139,7 @@ def test_anthropic_passthrough_tool_call_logs_cost(
 ) -> None:
     result = client.anthropic_message(
         scoped_key,
-        "claude-haiku-4-5",
+        ANTHROPIC_CHAT.alias,
         "What is the weather in Paris? Use the get_weather tool.",
         tools=[
             AnthropicTool(
diff --git a/tests/e2e/llm_translation/test_provider_features_e2e.py b/tests/e2e/llm_translation/test_provider_features_e2e.py
index cf05a4306b4..64df2abdadf 100644
--- a/tests/e2e/llm_translation/test_provider_features_e2e.py
+++ b/tests/e2e/llm_translation/test_provider_features_e2e.py
@@ -27,6 +27,7 @@ from pydantic import BaseModel
 from e2e_config import unique_marker
 from e2e_http import unwrap
 from lifecycle import ResourceManager
+from model_matrix import BEDROCK_ANTHROPIC_CHAT, OPENAI_CHAT
 from models import ChatBody, ChatMessage, ChatResponse, LiteLLMParamsBody
 from passthrough_client import PassthroughClient
 
@@ -86,7 +87,7 @@ class TestServiceTier:
         model_id = client.gateway.create_model(
             model,
             LiteLLMParamsBody(
-                model="openai/gpt-5.5", api_key="os.environ/OPENAI_API_KEY"
+                model=OPENAI_CHAT.backend, api_key="os.environ/OPENAI_API_KEY"
             ),
         )
         resources.defer(lambda: client.gateway.delete_model(model_id))
@@ -121,7 +122,7 @@ class TestPromptCaching:
         model_id = client.gateway.create_model(
             model,
             LiteLLMParamsBody(
-                model="bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0",
+                model=BEDROCK_ANTHROPIC_CHAT.backend,
                 aws_region_name="us-east-1",
             ),
         )
diff --git a/tests/e2e/llm_translation/test_rerank_e2e.py b/tests/e2e/llm_translation/test_rerank_e2e.py
index 4b30ac1ea5c..5866cc7d031 100644
--- a/tests/e2e/llm_translation/test_rerank_e2e.py
+++ b/tests/e2e/llm_translation/test_rerank_e2e.py
@@ -13,6 +13,7 @@ from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from endpoints_client import EndpointsClient, RerankResult
 from lifecycle import ResourceManager
+from model_matrix import COHERE_RERANK
 from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
@@ -32,7 +33,7 @@ class TestRerank:
         model = f"e2e-rerank-{unique_marker()}"
         model_id = endpoints_client.create_model(
             model,
-            LiteLLMParamsBody(model="cohere/rerank-v3.5", api_key="os.environ/COHERE_API_KEY"),
+            LiteLLMParamsBody(model=COHERE_RERANK.backend, api_key="os.environ/COHERE_API_KEY"),
         )
         resources.defer(lambda: endpoints_client.delete_model(model_id))
         key = resources.key()
diff --git a/tests/e2e/llm_translation/test_responses_e2e.py b/tests/e2e/llm_translation/test_responses_e2e.py
index 743de79880f..350d843fe90 100644
--- a/tests/e2e/llm_translation/test_responses_e2e.py
+++ b/tests/e2e/llm_translation/test_responses_e2e.py
@@ -13,6 +13,7 @@ from e2e_config import unique_marker
 from e2e_http import require_successful_call
 from endpoints_client import EndpointsClient, ResponsesResult
 from lifecycle import ResourceManager
+from model_matrix import OPENAI_CHAT_MINI
 from models import LiteLLMParamsBody
 
 pytestmark = pytest.mark.e2e
@@ -25,7 +26,7 @@ class TestResponses:
         model = f"e2e-responses-{unique_marker()}"
         model_id = endpoints_client.create_model(
             model,
-            LiteLLMParamsBody(model="openai/gpt-4o-mini", api_key="os.environ/OPENAI_API_KEY"),
+            LiteLLMParamsBody(model=OPENAI_CHAT_MINI.backend, api_key="os.environ/OPENAI_API_KEY"),
         )
         resources.defer(lambda: endpoints_client.delete_model(model_id))
         key = resources.key()
diff --git a/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py b/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
index 78d2bb358d2..5bcab1a2cfa 100644
--- a/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
+++ b/tests/e2e/llm_translation/test_vertex_passthrough_e2e.py
@@ -30,12 +30,13 @@ from pydantic import BaseModel
 from e2e_config import unique_marker
 from e2e_http import NoBody, require_successful_call, unwrap
 from lifecycle import ResourceManager
+from model_matrix import GEMINI_CHAT
 from models import SpendLogRow
 from passthrough_client import PassthroughClient
 
 pytestmark = pytest.mark.e2e
 
-VERTEX_MODEL = "gemini-2.5-flash"
+VERTEX_MODEL = GEMINI_CHAT.alias
 # The added deployment's region and the passthrough URL's region are the same constant,
 # so they always agree; the proxy registers passthrough credentials per project+region.
 VERTEX_LOCATION = os.environ.get("VERTEXAI_LOCATION", "us-central1")
diff --git a/tests/e2e/logging/test_prometheus_cardinality_e2e.py b/tests/e2e/logging/test_prometheus_cardinality_e2e.py
index 163293a3009..d1d5599fc08 100644
--- a/tests/e2e/logging/test_prometheus_cardinality_e2e.py
+++ b/tests/e2e/logging/test_prometheus_cardinality_e2e.py
@@ -23,10 +23,11 @@ from prometheus_client.parser import text_string_to_metric_families
 from e2e_config import unique_marker
 from lifecycle import ResourceManager
 from logging_client import LoggingClient
+from model_matrix import GEMINI_CHAT
 
 pytestmark = pytest.mark.e2e
 
-DRIVER_MODEL = "gemini-2.5-flash"
+DRIVER_MODEL = GEMINI_CHAT.alias
 REQUESTS_METRIC = "litellm_requests_metric_total"
 ALIAS_LABEL = "api_key_alias"
 DISTINCT_KEYS = 3
diff --git a/tests/e2e/management/test_key_models_dropdown_e2e.py b/tests/e2e/management/test_key_models_dropdown_e2e.py
index f0ba21699e0..8234e453499 100644
--- a/tests/e2e/management/test_key_models_dropdown_e2e.py
+++ b/tests/e2e/management/test_key_models_dropdown_e2e.py
@@ -17,6 +17,7 @@ import pytest
 from e2e_config import PROXY_BASE_URL, unique_marker
 from lifecycle import ResourceManager
 from management_client import ManagementClient
+from model_matrix import OPENAI_CHAT
 from models import KeyGenerateBody, TeamNewBody
 
 pytest.importorskip("playwright.sync_api", reason="playwright not installed")
@@ -77,7 +78,7 @@ def _open_key_edit_form(page: Page, key_alias: str) -> None:
 
 
 def _provision_team(client: ManagementClient, resources: ResourceManager, alias: str) -> str:
-    team_id = client.create_team(TeamNewBody(team_alias=alias, models=["all-proxy-models", "gpt-5.5"]))
+    team_id = client.create_team(TeamNewBody(team_alias=alias, models=["all-proxy-models", OPENAI_CHAT.alias]))
     resources.defer(lambda: client.delete_team(team_id))
     return team_id
 
@@ -85,7 +86,7 @@ def _provision_team(client: ManagementClient, resources: ResourceManager, alias:
 def _provision_key(
     client: ManagementClient, resources: ResourceManager, alias: str, team_id: str | None = None
 ) -> str:
-    key = client.gateway.generate_key(KeyGenerateBody(key_alias=alias, models=["gpt-5.5"], team_id=team_id))
+    key = client.gateway.generate_key(KeyGenerateBody(key_alias=alias, models=[OPENAI_CHAT.alias], team_id=team_id))
     resources.defer(lambda: client.gateway.delete_key(key))
     return key
 
@@ -98,7 +99,7 @@ class TestKeyModelsDropdownUI:
     ) -> None:
         _open_create_key_modal(ui_page)
 
-        options = _models_dropdown_texts(ui_page, must_contain="gpt-5.5")
+        options = _models_dropdown_texts(ui_page, must_contain=OPENAI_CHAT.alias)
         assert "All Proxy Models" in options, f"teamless create lost 'All Proxy Models': {options}"
         assert "All Team Models" not in options, f"teamless create offered 'All Team Models': {options}"
 
@@ -120,7 +121,7 @@ class TestKeyModelsDropdownUI:
         _select_team(ui_page, team_alias)
 
         options = _models_dropdown_texts(ui_page, must_contain="All Team Models")
-        assert "gpt-5.5" in options, f"team key create lost the team's own model: {options}"
+        assert OPENAI_CHAT.alias in options, f"team key create lost the team's own model: {options}"
         assert "All Proxy Models" not in options, f"team key create offered 'All Proxy Models': {options}"
         assert "all-proxy-models" not in options, f"team key create offered the raw sentinel: {options}"
 
@@ -140,7 +141,7 @@ class TestKeyModelsDropdownUI:
 
         _open_key_edit_form(ui_page, key_alias)
 
-        options = _models_dropdown_texts(ui_page, must_contain="gpt-5.5")
+        options = _models_dropdown_texts(ui_page, must_contain=OPENAI_CHAT.alias)
         assert "All Proxy Models" in options, f"teamless edit lost 'All Proxy Models': {options}"
         assert "All Team Models" not in options, f"teamless edit offered 'All Team Models': {options}"
 
@@ -156,6 +157,6 @@ class TestKeyModelsDropdownUI:
         _open_key_edit_form(ui_page, key_alias)
 
         options = _models_dropdown_texts(ui_page, must_contain="All Team Models")
-        assert "gpt-5.5" in options, f"team key edit lost the team's own model: {options}"
+        assert OPENAI_CHAT.alias in options, f"team key edit lost the team's own model: {options}"
         assert "All Proxy Models" not in options, f"team key edit offered 'All Proxy Models': {options}"
         assert "all-proxy-models" not in options, f"team key edit offered the raw sentinel: {options}"
diff --git a/tests/e2e/management/test_management_e2e.py b/tests/e2e/management/test_management_e2e.py
index cbd5db0d59f..3af5e4ded8f 100644
--- a/tests/e2e/management/test_management_e2e.py
+++ b/tests/e2e/management/test_management_e2e.py
@@ -22,6 +22,7 @@ from management_client import (
     ROUTE_NOT_ALLOWED_MARKER,
     ManagementClient,
 )
+from model_matrix import GEMINI_CHAT, OPENAI_CHAT
 from models import KeyGenerateBody, OrgNewBody, TeamNewBody, UserNewBody
 
 pytestmark = pytest.mark.e2e
@@ -111,42 +112,42 @@ class TestKeyRoutes:
         key = _generate_key(
             client,
             resources,
-            KeyGenerateBody(models=["gemini-2.5-flash"], key_alias=alias, tpm_limit=424242),
+            KeyGenerateBody(models=[GEMINI_CHAT.alias], key_alias=alias, tpm_limit=424242),
         )
 
         info = client.gateway.key_info(key)
         assert info.key_alias == alias, f"/key/info reports key_alias {info.key_alias!r}, configured {alias!r}"
-        assert info.models == ["gemini-2.5-flash"], (
-            f"/key/info reports models {info.models}, configured ['gemini-2.5-flash']"
+        assert info.models == [GEMINI_CHAT.alias], (
+            f"/key/info reports models {info.models}, configured [{GEMINI_CHAT.alias!r}]"
         )
         assert info.tpm_limit == 424242, (
             f"/key/info reports tpm_limit {info.tpm_limit}, configured 424242"
         )
 
-        _poll_chat_ok(client, key, "gemini-2.5-flash")
+        _poll_chat_ok(client, key, GEMINI_CHAT.alias)
         _assert_model_denied(
-            client.chat_status(key, "gpt-5.5", f"say hi {unique_marker()}"), "gpt-5.5"
+            client.chat_status(key, OPENAI_CHAT.alias, f"say hi {unique_marker()}"), OPENAI_CHAT.alias
         )
 
     @pytest.mark.covers("mgmt.key.update.persists")
     def test_update_models_persists_and_flips_enforcement(
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
-        key = _generate_key(client, resources, KeyGenerateBody(models=["gemini-2.5-flash"]))
-        _poll_chat_ok(client, key, "gemini-2.5-flash")
+        key = _generate_key(client, resources, KeyGenerateBody(models=[GEMINI_CHAT.alias]))
+        _poll_chat_ok(client, key, GEMINI_CHAT.alias)
         _assert_model_denied(
-            client.chat_status(key, "gpt-5.5", f"say hi {unique_marker()}"), "gpt-5.5"
+            client.chat_status(key, OPENAI_CHAT.alias, f"say hi {unique_marker()}"), OPENAI_CHAT.alias
         )
 
-        client.update_key_models(key, ["gpt-5.5"])
+        client.update_key_models(key, [OPENAI_CHAT.alias])
 
         info = client.gateway.key_info(key)
-        assert info.models == ["gpt-5.5"], (
-            f"/key/info reports models {info.models} after /key/update to ['gpt-5.5']"
+        assert info.models == [OPENAI_CHAT.alias], (
+            f"/key/info reports models {info.models} after /key/update to [{OPENAI_CHAT.alias!r}]"
         )
 
-        _poll_model_access_granted(client, key, "gpt-5.5")
-        _poll_chat_denied(client, key, "gemini-2.5-flash")
+        _poll_model_access_granted(client, key, OPENAI_CHAT.alias)
+        _poll_chat_denied(client, key, GEMINI_CHAT.alias)
 
     @pytest.mark.covers("mgmt.key.delete.persists")
     def test_delete_revokes_the_key_on_chat(self, client: ManagementClient, resources: ResourceManager) -> None:
@@ -154,13 +155,13 @@ class TestKeyRoutes:
         design: the deferred cleanup must survive this test failing before the
         in-body delete, and a repeat /key/delete is a cheap no-op the warn-only
         teardown absorbs."""
-        key = _generate_key(client, resources, KeyGenerateBody(models=["gemini-2.5-flash"]))
-        _poll_chat_ok(client, key, "gemini-2.5-flash")
+        key = _generate_key(client, resources, KeyGenerateBody(models=[GEMINI_CHAT.alias]))
+        _poll_chat_ok(client, key, GEMINI_CHAT.alias)
 
         client.delete_key_strict(key)
 
         def rejected() -> bool | None:
-            outcome = client.chat_status(key, "gemini-2.5-flash", f"say hi {unique_marker()}")
+            outcome = client.chat_status(key, GEMINI_CHAT.alias, f"say hi {unique_marker()}")
             return True if outcome.status_code == 401 else None
 
         _ = _poll(client, rejected, "deleted key was still accepted on chat (never rejected 401) at the deadline")
@@ -172,12 +173,12 @@ class TestTeamRoutes:
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
         alias = f"e2e-mgmt-team-{unique_marker()}"
-        team_id = _create_team(client, resources, alias, ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, alias, [GEMINI_CHAT.alias])
 
         info = client.team_info(team_id)
         assert info.team_alias == alias, f"/team/info reports team_alias {info.team_alias!r}, configured {alias!r}"
-        assert info.models == ["gemini-2.5-flash"], (
-            f"/team/info reports models {info.models}, configured ['gemini-2.5-flash']"
+        assert info.models == [GEMINI_CHAT.alias], (
+            f"/team/info reports models {info.models}, configured [{GEMINI_CHAT.alias!r}]"
         )
 
         key = _generate_key(client, resources, KeyGenerateBody(team_id=team_id))
@@ -195,7 +196,7 @@ class TestTeamRoutes:
             resources,
             UserNewBody(user_email=f"e2e-mgmt-{unique_marker()}@example.com", user_role="internal_user"),
         )
-        team_id = _create_team(client, resources, f"e2e-mgmt-team-{unique_marker()}", ["gemini-2.5-flash"])
+        team_id = _create_team(client, resources, f"e2e-mgmt-team-{unique_marker()}", [GEMINI_CHAT.alias])
 
         client.add_team_member(team_id, user_id)
         member = next(
@@ -230,15 +231,15 @@ class TestOrganizationRoutes:
         self, client: ManagementClient, resources: ResourceManager
     ) -> None:
         alias = f"e2e-mgmt-org-{unique_marker()}"
-        org_id = client.create_org(OrgNewBody(organization_alias=alias, models=["gemini-2.5-flash"]))
+        org_id = client.create_org(OrgNewBody(organization_alias=alias, models=[GEMINI_CHAT.alias]))
         resources.defer(lambda: client.delete_org(org_id))
 
         info = client.org_info(org_id)
         assert info.organization_alias == alias, (
             f"/organization/info reports alias {info.organization_alias!r}, configured {alias!r}"
         )
-        assert info.models == ["gemini-2.5-flash"], (
-            f"/organization/info reports models {info.models}, configured ['gemini-2.5-flash']"
+        assert info.models == [GEMINI_CHAT.alias], (
+            f"/organization/info reports models {info.models}, configured [{GEMINI_CHAT.alias!r}]"
         )
 
 
diff --git a/tests/e2e/model_matrix.py b/tests/e2e/model_matrix.py
new file mode 100644
index 00000000000..1c59b63ac09
--- /dev/null
+++ b/tests/e2e/model_matrix.py
@@ -0,0 +1,66 @@
+"""Single source of truth for every model the e2e suite drives.
+
+Bump a model version here instead of editing individual tests. Constants are
+named for the role a model plays, never its version, so a bump touches this
+file (plus docker-compose.yml, which cannot import Python) and nothing else.
+tests/code_coverage_tests/check_e2e_model_freshness.py fails CI when a pin
+disappears from model_prices_and_context_window.json, approaches its
+deprecation_date, drifts from the docker-compose gateway config, or when a
+test hardcodes a model literal instead of importing a pin.
+"""
+
+from dataclasses import dataclass
+
+
+@dataclass(frozen=True, slots=True)
+class ModelPin:
+    provider: str
+    model_id: str
+    gateway_alias: str = ""
+    pricing_key: str = ""
+
+    @property
+    def backend(self) -> str:
+        return f"{self.provider}/{self.model_id}"
+
+    @property
+    def alias(self) -> str:
+        return self.gateway_alias or self.model_id
+
+    @property
+    def canonical(self) -> str:
+        return self.pricing_key or self.backend
+
+
+GEMINI_CHAT = ModelPin("gemini", "gemini-3.5-flash")
+OPENAI_CHAT = ModelPin("openai", "gpt-5.5")
+OPENAI_CHAT_MINI = ModelPin("openai", "gpt-5.4-mini")
+ANTHROPIC_CHAT = ModelPin("anthropic", "claude-haiku-4-5")
+OPENAI_EMBEDDING = ModelPin(
+    "openai", "text-embedding-3-small", gateway_alias="openai-text-embedding-3-small"
+)
+OPENAI_TTS = ModelPin("openai", "gpt-4o-mini-tts")
+VERTEX_CHAT = ModelPin("vertex_ai", "gemini-3.5-flash")
+AZURE_BATCH = ModelPin("azure", "gpt-4.1-mini-batch", pricing_key="azure/gpt-4.1-mini")
+BEDROCK_ANTHROPIC_CHAT = ModelPin("bedrock", "us.anthropic.claude-haiku-4-5-20251001-v1:0")
+COHERE_RERANK = ModelPin("cohere", "rerank-v3.5")
+
+GATEWAY_MODELS: tuple[ModelPin, ...] = (
+    OPENAI_CHAT,
+    ANTHROPIC_CHAT,
+    GEMINI_CHAT,
+    OPENAI_EMBEDDING,
+)
+
+ALL_PINS: tuple[ModelPin, ...] = (
+    GEMINI_CHAT,
+    OPENAI_CHAT,
+    OPENAI_CHAT_MINI,
+    ANTHROPIC_CHAT,
+    OPENAI_EMBEDDING,
+    OPENAI_TTS,
+    VERTEX_CHAT,
+    AZURE_BATCH,
+    BEDROCK_ANTHROPIC_CHAT,
+    COHERE_RERANK,
+)
diff --git a/tests/e2e/spend_tracking/conftest.py b/tests/e2e/spend_tracking/conftest.py
index 0e80764236b..a03c72b640f 100644
--- a/tests/e2e/spend_tracking/conftest.py
+++ b/tests/e2e/spend_tracking/conftest.py
@@ -19,6 +19,7 @@ from typing import Iterator
 
 import pytest
 
+from model_matrix import ANTHROPIC_CHAT, GEMINI_CHAT, OPENAI_EMBEDDING
 from models import LiteLLMParamsBody
 from spend_e2e_client import SpendClient, build_client
 
@@ -31,9 +32,9 @@ def _driver_params(provider_model: str, env_var: str) -> LiteLLMParamsBody:
 
 
 DRIVER_MODELS: tuple[tuple[str, str, str], ...] = (
-    ("gemini-2.5-flash", "gemini/gemini-2.5-flash", "GEMINI_API_KEY"),
-    ("claude-haiku-4-5", "anthropic/claude-haiku-4-5", "ANTHROPIC_API_KEY"),
-    ("openai-text-embedding-3-small", "openai/text-embedding-3-small", "OPENAI_API_KEY"),
+    (GEMINI_CHAT.alias, GEMINI_CHAT.backend, "GEMINI_API_KEY"),
+    (ANTHROPIC_CHAT.alias, ANTHROPIC_CHAT.backend, "ANTHROPIC_API_KEY"),
+    (OPENAI_EMBEDDING.alias, OPENAI_EMBEDDING.backend, "OPENAI_API_KEY"),
 )
 
 
diff --git a/tests/e2e/spend_tracking/test_spend_tracking_e2e.py b/tests/e2e/spend_tracking/test_spend_tracking_e2e.py
index 3495011eab6..75eceebb39c 100644
--- a/tests/e2e/spend_tracking/test_spend_tracking_e2e.py
+++ b/tests/e2e/spend_tracking/test_spend_tracking_e2e.py
@@ -3,8 +3,8 @@
 Run against a proxy started with the gateway config. Coverage rationale:
 SPEND_TRACKING_COVERAGE_MATRIX.md.
 
-Model names are literals from that config: chat tests hit "gemini-2.5-flash",
-embedding tests hit "openai-text-embedding-3-small".
+Model names come from model_matrix.py pins baked into that config: chat tests
+hit GEMINI_CHAT, embedding tests hit OPENAI_EMBEDDING.
 
 Every test: fresh scoped key (isolation) -> real provider call -> unwrap (hard
 fail if the proxy couldn't make a call it should) -> poll /spend/logs to a
@@ -23,6 +23,7 @@ import pytest
 
 from e2e_http import Result, Success
 from lifecycle import ResourceManager
+from model_matrix import ANTHROPIC_CHAT, GEMINI_CHAT, OPENAI_EMBEDDING
 from models import ChatResponse, SpendLogs, SpendLogsParams
 from spend_e2e_client import SpendClient, SpendLogRow, is_ok, unique_marker, unwrap
 
@@ -65,7 +66,7 @@ def test_chat_completion_writes_nonzero_spend_row(
     chat = unwrap(
         client.chat(
             scoped_key,
-            "gemini-2.5-flash",
+            GEMINI_CHAT.alias,
             f"reply with one word {unique_marker()}",
             max_tokens=16,
         )
@@ -79,7 +80,7 @@ def test_chat_completion_writes_nonzero_spend_row(
     assert (row.spend or 0) > 0, f"chat row should cost > 0: {_summarize(rows)}"
     assert row.status == "success"
     assert row.cache_hit != "True", "fresh call must not be a cache hit"
-    assert "gemini-2.5-flash" in (row.model or "")
+    assert GEMINI_CHAT.alias in (row.model or "")
 
     prompt = row.prompt_tokens or 0
     completion = row.completion_tokens or 0
@@ -98,7 +99,7 @@ def test_streaming_chat_completion_tracks_spend(
 ) -> None:
     result = client.chat_stream(
         scoped_key,
-        "gemini-2.5-flash",
+        GEMINI_CHAT.alias,
         f"count to three {unique_marker()}",
         max_tokens=64,
     )
@@ -126,7 +127,7 @@ def test_embedding_writes_nonzero_spend_row(
     _ = unwrap(
         client.embed(
             scoped_key,
-            "openai-text-embedding-3-small",
+            OPENAI_EMBEDDING.alias,
             f"vectorize this sentence {unique_marker()}",
         )
     )
@@ -139,7 +140,7 @@ def test_embedding_writes_nonzero_spend_row(
     )
     assert (row.prompt_tokens or 0) > 0
     assert (row.completion_tokens or 0) == 0, "embeddings have no completion tokens"
-    assert "text-embedding-3-small" in (row.model or "")
+    assert OPENAI_EMBEDDING.model_id in (row.model or "")
 
 
 def test_cache_hit_is_zero_cost_and_suffixed(
@@ -150,8 +151,8 @@ def test_cache_hit_is_zero_cost_and_suffixed(
     # populated. The marker keeps each run isolated - a fixed prompt would persist
     # in the shared response cache across runs and make both calls hit (flaky).
     prompt = f"What is the capital of France? Answer in one word. {unique_marker()}"
-    _ = unwrap(client.chat(scoped_key, "gemini-2.5-flash", prompt, max_tokens=16))
-    _ = unwrap(client.chat(scoped_key, "gemini-2.5-flash", prompt, max_tokens=16))
+    _ = unwrap(client.chat(scoped_key, GEMINI_CHAT.alias, prompt, max_tokens=16))
+    _ = unwrap(client.chat(scoped_key, GEMINI_CHAT.alias, prompt, max_tokens=16))
 
     rows = client.poll_logs_for_key(
         scoped_key, predicate=lambda rs: any(r.cache_hit == "True" for r in rs)
@@ -182,7 +183,7 @@ def test_key_spend_equals_sum_of_logs(client: SpendClient, scoped_key: str) -> N
         _ = unwrap(
             client.chat(
                 scoped_key,
-                "gemini-2.5-flash",
+                GEMINI_CHAT.alias,
                 f"say hi {unique_marker()}",
                 max_tokens=16,
             )
@@ -216,7 +217,7 @@ def test_burst_of_concurrent_calls_loses_no_spend(
     def call(idx: int) -> Result[ChatResponse]:
         return client.chat(
             scoped_key,
-            "gemini-2.5-flash",
+            GEMINI_CHAT.alias,
             f"burst call {idx} {unique_marker()}",
             max_tokens=16,
         )
@@ -265,7 +266,7 @@ def test_spend_logs_v2_pagination_caps_pages_and_keeps_total(
         _ = unwrap(
             client.chat(
                 scoped_key,
-                "gemini-2.5-flash",
+                GEMINI_CHAT.alias,
                 f"page fodder {unique_marker()}",
                 max_tokens=16,
             )
@@ -305,7 +306,7 @@ def test_request_tags_round_trip(client: SpendClient, scoped_key: str) -> None:
     tag = f"e2e-spend-{unique_marker()}"
     _ = unwrap(
         client.chat(
-            scoped_key, "gemini-2.5-flash", "tagged request", tags=[tag], max_tokens=16
+            scoped_key, GEMINI_CHAT.alias, "tagged request", tags=[tag], max_tokens=16
         )
     )
 
@@ -327,7 +328,7 @@ def test_tag_spend_matches_sum_of_tagged_logs(
         _ = unwrap(
             client.chat(
                 scoped_key,
-                "gemini-2.5-flash",
+                GEMINI_CHAT.alias,
                 f"hi {unique_marker()}",
                 tags=[tag],
                 max_tokens=16,
@@ -359,7 +360,7 @@ def test_end_user_spend_attributed_on_row(
 ) -> None:
     customer = resources.customer(f"e2e-cust-{unique_marker()}")
     _ = unwrap(
-        client.chat(scoped_key, "gemini-2.5-flash", "hi", user=customer, max_tokens=16)
+        client.chat(scoped_key, GEMINI_CHAT.alias, "hi", user=customer, max_tokens=16)
     )
 
     rows = client.poll_logs_for_key(
@@ -381,27 +382,27 @@ def test_each_model_on_a_shared_key_gets_its_own_row(
     sibling deployment, or collapses both calls onto one request_id fails here."""
     gemini = unwrap(
         client.chat(
-            scoped_key, "gemini-2.5-flash", f"one word {unique_marker()}", max_tokens=16
+            scoped_key, GEMINI_CHAT.alias, f"one word {unique_marker()}", max_tokens=16
         )
     )
     claude = unwrap(
         client.chat(
-            scoped_key, "claude-haiku-4-5", f"one word {unique_marker()}", max_tokens=16
+            scoped_key, ANTHROPIC_CHAT.alias, f"one word {unique_marker()}", max_tokens=16
         )
     )
 
     def both_models_costed(rows: list[SpendLogRow]) -> bool:
         costed = [r.model or "" for r in rows if (r.spend or 0) > 0]
-        return any("gemini-2.5-flash" in m for m in costed) and any(
-            "claude-haiku-4-5" in m for m in costed
+        return any(GEMINI_CHAT.alias in m for m in costed) and any(
+            ANTHROPIC_CHAT.alias in m for m in costed
         )
 
     rows = client.poll_logs_for_key(scoped_key, min_rows=2, predicate=both_models_costed)
     gemini_row = _require_row(
-        rows, lambda r: "gemini-2.5-flash" in (r.model or ""), "for the gemini call"
+        rows, lambda r: GEMINI_CHAT.alias in (r.model or ""), "for the gemini call"
     )
     claude_row = _require_row(
-        rows, lambda r: "claude-haiku-4-5" in (r.model or ""), "for the claude call"
+        rows, lambda r: ANTHROPIC_CHAT.alias in (r.model or ""), "for the claude call"
     )
 
     assert (gemini_row.spend or 0) > 0, f"gemini row should cost > 0: {_summarize(rows)}"
@@ -422,7 +423,7 @@ def test_each_model_on_a_shared_key_gets_its_own_row(
 def test_failure_call_writes_failure_status_row(
     client: SpendClient, scoped_key: str
 ) -> None:
-    result = client.chat(scoped_key, "gemini-2.5-flash", "", max_tokens=1)
+    result = client.chat(scoped_key, GEMINI_CHAT.alias, "", max_tokens=1)
     if is_ok(result):
         pytest.skip("call unexpectedly succeeded; could not induce a failure row")
 
@@ -440,11 +441,11 @@ def test_failure_call_writes_failure_status_row(
 
 def test_spend_calculate_returns_nonzero_cost(client: SpendClient) -> None:
     cost = client.calculate_spend(
-        "gemini-2.5-flash", "estimate the cost of this request"
+        GEMINI_CHAT.alias, "estimate the cost of this request"
     )
     assert cost > 0, (
-        "/spend/calculate returned 0 for gemini-2.5-flash; "
-        "cost map may be missing this model"
+        f"/spend/calculate returned 0 for {GEMINI_CHAT.alias}; "
+        f"cost map may be missing this model"
     )
 
 
@@ -458,7 +459,7 @@ def test_spend_logs_endpoint_returns_spend(
     call's nonzero spend must surface before the deadline."""
     unwrap(
         client.chat(
-            scoped_key, "gemini-2.5-flash", f"spend logs {unique_marker()}", max_tokens=16
+            scoped_key, GEMINI_CHAT.alias, f"spend logs {unique_marker()}", max_tokens=16
         )
     )
 
diff --git a/tests/e2e/test_e2e_gateway.py b/tests/e2e/test_e2e_gateway.py
index a6dcc6112d6..ceb64121aec 100644
--- a/tests/e2e/test_e2e_gateway.py
+++ b/tests/e2e/test_e2e_gateway.py
@@ -22,6 +22,7 @@ from e2e_http import (
     StreamingResponse,
     Success,
 )
+from model_matrix import OPENAI_CHAT_MINI
 from models import (
     LiteLLMParamsBody,
     ModelDeleteBody,
@@ -109,7 +110,7 @@ def test_gateway_create_model_registers_deployment_and_returns_model_id() -> Non
     gateway = Gateway(transport=transport)
 
     model_id = gateway.create_model(
-        "e2e-test-model", LiteLLMParamsBody(model="openai/gpt-4o-mini")
+        "e2e-test-model", LiteLLMParamsBody(model=OPENAI_CHAT_MINI.backend)
     )
 
     assert model_id == "registered-id"
@@ -126,7 +127,7 @@ def test_batch_client_create_model_registers_a_batch_mode_deployment() -> None:
     client = BatchClient(gateway=Gateway(transport=transport))
 
     model_id = client.create_model(
-        "e2e-batch-model", LiteLLMParamsBody(model="openai/gpt-4o-mini")
+        "e2e-batch-model", LiteLLMParamsBody(model=OPENAI_CHAT_MINI.backend)
     )
 
     assert model_id == "registered-id"