refactor(anthropic): rename experimental_pass_through to pass_through (#43329)

* refactor(anthropic): rename experimental_pass_through to pass_through

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

* test(anthropic): point compact patch targets at renamed pass_through path

Co-Authored-By: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

---------

Co-authored-by: kerry <kerry@berri.ai>
Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>

ARCHITECTURE.md

@@ -336,7 +336,7 @@ Each translation is isolated in its own file, making it easy to test and modify
 | `/v1/chat/completions` | Gemini | `llms/gemini/chat/transformation.py` |
 | `/v1/chat/completions` | Vertex AI | `llms/vertex_ai/gemini/transformation.py` |
 | `/v1/chat/completions` | OpenAI | `llms/openai/chat/gpt_transformation.py` |
-| `/v1/messages` (passthrough) | Anthropic | `llms/anthropic/experimental_pass_through/messages/transformation.py` |
+| `/v1/messages` (passthrough) | Anthropic | `llms/anthropic/pass_through/messages/transformation.py` |
 | `/v1/messages` (passthrough) | Bedrock | `llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py` |
 | `/v1/messages` (passthrough) | Vertex AI | `llms/vertex_ai/vertex_ai_partner_models/anthropic/experimental_pass_through/transformation.py` |
 | Passthrough endpoints | All | `proxy/pass_through_endpoints/llm_provider_handlers/` |

litellm/__init__.py

@@ -1691,7 +1691,7 @@ if TYPE_CHECKING:
         SagemakerNovaConfig as SagemakerNovaConfig,
     )
     from .llms.cohere.chat.transformation import CohereChatConfig as CohereChatConfig
-    from .llms.anthropic.experimental_pass_through.messages.transformation import (
+    from .llms.anthropic.pass_through.messages.transformation import (
         AnthropicMessagesConfig as AnthropicMessagesConfig,
     )
     from .llms.bedrock.messages.invoke_transformations.anthropic_claude3_transformation import (

litellm/_lazy_imports_registry.py

@@ -742,7 +742,7 @@ _LLM_CONFIGS_IMPORT_MAP: Final = {
     ),
     "CohereChatConfig": (".llms.cohere.chat.transformation", "CohereChatConfig"),
     "AnthropicMessagesConfig": (
-        ".llms.anthropic.experimental_pass_through.messages.transformation",
+        ".llms.anthropic.pass_through.messages.transformation",
         "AnthropicMessagesConfig",
     ),
     "BedrockClaudePlatformMessagesConfig": (

litellm/caching/caching_handler.py

@@ -52,7 +52,7 @@ from litellm.types.utils import (
 
 if TYPE_CHECKING:
     from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
-    from litellm.llms.anthropic.experimental_pass_through.messages.response_cache import (
+    from litellm.llms.anthropic.pass_through.messages.response_cache import (
         AnthropicMessagesStreamCacheWriter,
     )
     from litellm.types.utils import PromptTokensDetailsWrapper
@@ -127,7 +127,7 @@ def _should_defer_streaming_cache_hit_callbacks(*, cached_result: object) -> boo
     spend and callback records. A plain (non-stream) replay logs here, since nothing
     else will.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.response_cache import (
+    from litellm.llms.anthropic.pass_through.messages.response_cache import (
         CachedAnthropicMessagesStreamIterator,
     )
     from litellm.responses.streaming_iterator import BaseResponsesAPIStreamingIterator
@@ -930,7 +930,7 @@ class LLMCachingHandler:
         elif (
             call_type == CallTypes.anthropic_messages.value or call_type == CallTypes.aanthropic_messages.value
         ) and isinstance(cached_result, dict):
-            from litellm.llms.anthropic.experimental_pass_through.messages.response_cache import (
+            from litellm.llms.anthropic.pass_through.messages.response_cache import (
                 convert_cached_anthropic_messages_result,
             )
 
@@ -1150,7 +1150,7 @@ class LLMCachingHandler:
             return result
         if not isinstance(result, AsyncIterator):
             return result
-        from litellm.llms.anthropic.experimental_pass_through.messages.response_cache import (
+        from litellm.llms.anthropic.pass_through.messages.response_cache import (
             AnthropicMessagesStreamCacheWriter,
         )
 

litellm/integrations/shadow_eval_logger.py

@@ -111,7 +111,7 @@ def _chat_request_from_anthropic_messages(
     because the logged optional_params switch dialect per provider path (the bridge's
     inner completion rewrites them to chat shape mid-flight); the adapter translates
     them alongside the messages, and sampling params copy through untranslated."""
-    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         LiteLLMAnthropicMessagesAdapter,
     )
 

litellm/integrations/websearch_interception/ARCHITECTURE.md

@@ -70,7 +70,7 @@ Claude Code (Anthropic's official CLI) sends web search requests using Anthropic
 
 Native tools are converted to LiteLLM standard format **before** sending to the provider:
 
-1. **Conversion Point** (`litellm/llms/anthropic/experimental_pass_through/messages/handler.py`):
+1. **Conversion Point** (`litellm/llms/anthropic/pass_through/messages/handler.py`):
    - In `anthropic_messages()` function (lines 60-127)
    - Runs BEFORE the API request is made
    - Detects native web search tools using `is_web_search_tool()`

litellm/llms/anthropic/chat/guardrail_translation/handler.py

@@ -25,7 +25,7 @@ from typing_extensions import ReadOnly, TypedDict, assert_never
 
 from litellm._logging import verbose_proxy_logger
 from litellm.llms.anthropic.chat.transformation import AnthropicConfig
-from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+from litellm.llms.anthropic.pass_through.adapters.transformation import (
     LiteLLMAnthropicMessagesAdapter,
     is_provider_native_tool_dict,
 )
@@ -365,7 +365,7 @@ class AnthropicMessagesHandler(BaseTranslation):
     def _standalone_block_chunks(self, exc: "ModifyResponseException") -> list[bytes]:
         import uuid
 
-        from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
             FakeAnthropicMessagesStreamIterator,
         )
         from litellm.llms.base_llm.guardrail_translation.utils import (

litellm/llms/anthropic/pass_through/adapters/handler.py

@@ -11,15 +11,15 @@ from typing_extensions import TypedDict
 import litellm
 from litellm._logging import verbose_logger
 from litellm.litellm_core_utils.asyncify import run_async_function
-from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+from litellm.llms.anthropic.pass_through.adapters.transformation import (
     AnthropicAdapter,
 )
-from litellm.llms.anthropic.experimental_pass_through.context_management import (
+from litellm.llms.anthropic.pass_through.context_management import (
     AnthropicContextManagementError,
     PolyfillResult,
     apply_context_management,
 )
-from litellm.llms.anthropic.experimental_pass_through.utils import (
+from litellm.llms.anthropic.pass_through.utils import (
     is_reasoning_auto_summary_enabled,
     litellm_logging_obj_from_kwargs,
     local_model_name,
@@ -102,7 +102,7 @@ async def _prepare_context_managed_request(
     user_api_key_auth: "UserAPIKeyAuth | None" = None,
 ) -> PolyfillResult | None:
     """Apply client compaction history, then optional context_management polyfill."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         apply_client_compaction_block_history,
     )
 
@@ -179,7 +179,7 @@ def _polyfill_will_run(
     if edits is None:
         return False
 
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_EDIT_TYPE,
     )
 
@@ -205,7 +205,7 @@ def _spec_has_non_compact_edits(
     if edits is None:
         return False
 
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_EDIT_TYPE,
     )
 
@@ -240,7 +240,7 @@ def _normalize_spec_edits(
     if _context_management_explicitly_dropped(additional_drop_params):
         return None
 
-    from litellm.llms.anthropic.experimental_pass_through.context_management.dispatcher import (
+    from litellm.llms.anthropic.pass_through.context_management.dispatcher import (
         _normalize_spec,
     )
 
@@ -437,7 +437,7 @@ class LiteLLMMessagesToCompletionTransformationHandler:
         Handles both string ("max") and dict ({"effort": "max", "summary": ...})
         formats. Uses model registry to check supports_xhigh/supports_minimal.
         """
-        from litellm.llms.anthropic.experimental_pass_through.utils import (
+        from litellm.llms.anthropic.pass_through.utils import (
             normalize_reasoning_effort_value,
         )
 

litellm/llms/anthropic/pass_through/adapters/streaming_iterator.py

@@ -118,7 +118,7 @@ class _CombinedChunkSplitter:
     @staticmethod
     def _is_combined(chunk: "ModelResponseStream") -> bool:
         """True if ``chunk`` carries response content AND a finish_reason."""
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             openai_chat_refusal_text,
         )
 
@@ -1029,7 +1029,7 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
         delta: Final = processed_chunk["delta"]
         if delta.get("stop_reason") == "max_tokens":
             return processed_chunk
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             refusal_stop_details,
         )
 
@@ -1083,7 +1083,7 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
     @staticmethod
     def _is_blank_delta(chunk: "ModelResponseStream") -> bool:
         from litellm.llms.anthropic.common_utils import is_empty_unsigned_thinking_block
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             openai_chat_refusal_text,
         )
 
@@ -1120,7 +1120,7 @@ class AnthropicStreamWrapper(AdapterCompletionStreamWrapper):
         - Different content types in the response
         - Specific markers in the content
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             openai_chat_refusal_text,
         )
 

litellm/llms/anthropic/pass_through/adapters/transformation.py

@@ -8,7 +8,7 @@ from typing import TYPE_CHECKING, Any, Final, Literal, TypeAlias, TypeVar, cast
 from pydantic import JsonValue, TypeAdapter
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.utils import (
+from litellm.llms.anthropic.pass_through.utils import (
     is_reasoning_auto_summary_enabled,
     prompt_cache_key_from_user_id,
 )
@@ -134,14 +134,14 @@ from litellm.llms.anthropic.common_utils import (
     normalize_anthropic_tool_use_id,
     strip_encrypted_reasoning_blocks_from_anthropic_messages,
 )
-from litellm.llms.anthropic.experimental_pass_through.context_management import (
+from litellm.llms.anthropic.pass_through.context_management import (
     PolyfillResult,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.mid_conversation_system import (
+from litellm.llms.anthropic.pass_through.messages.mid_conversation_system import (
     convert_mid_conversation_system_turns,
     is_system_role_message,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+from litellm.llms.anthropic.pass_through.messages.utils import (
     openai_chat_refusal_text,
     refusal_stop_details,
 )

litellm/llms/anthropic/pass_through/context_management/editors/compact.py

@@ -798,7 +798,7 @@ def _count_effective_tokens(
     threshold check matches the downstream ``input_tokens`` metric.
     """
     # Local import to avoid pulling the adapter at module load time.
-    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         LiteLLMAnthropicMessagesAdapter,
     )
 
@@ -955,7 +955,7 @@ def _build_summary_messages(
     system prompt); the conversation history is translated to OpenAI shape;
     the summarization prompt is appended as a final user turn.
     """
-    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         LiteLLMAnthropicMessagesAdapter,
     )
 

litellm/llms/anthropic/pass_through/messages/agentic_streaming_iterator.py

@@ -336,7 +336,7 @@ class AgenticAnthropicStreamingIterator:
             await task
 
     async def aclose(self) -> None:
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             aclose_if_supported,
         )
 
@@ -379,7 +379,7 @@ class AgenticAnthropicStreamingIterator:
             if hasattr(result, "__aiter__"):
                 self._follow_up_iterator = result.__aiter__()
             elif isinstance(result, dict):
-                from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+                from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
                     FakeAnthropicMessagesStreamIterator,
                 )
                 from litellm.types.llms.anthropic_messages.anthropic_response import (

litellm/llms/anthropic/pass_through/messages/handler.py

@@ -215,7 +215,7 @@ async def _try_websearch_short_circuit(
         if response is not None:
             anthropic_response = cast(AnthropicMessagesResponse, response)
             if stream:
-                from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+                from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
                     FakeAnthropicMessagesStreamIterator,
                 )
 
@@ -531,7 +531,7 @@ def anthropic_messages_handler(
     # reference the provider cannot resolve. Popped from kwargs so it never reaches the provider.
     skip_mcp_handler: Final = kwargs.pop("_skip_mcp_handler", False)
     if not skip_mcp_handler and tools:
-        from litellm.llms.anthropic.experimental_pass_through.messages.mcp_handler import (
+        from litellm.llms.anthropic.pass_through.messages.mcp_handler import (
             anthropic_messages_with_mcp,
         )
         from litellm.responses.mcp.litellm_proxy_mcp_handler import (

litellm/llms/anthropic/pass_through/messages/interceptors/advisor.py

@@ -67,7 +67,7 @@ class AdvisorOrchestrationHandler(MessagesInterceptor):
         custom_llm_provider: str | None,
         **kwargs,
     ) -> AnthropicMessagesResponse | AsyncIterator:
-        from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
             FakeAnthropicMessagesStreamIterator,
         )
 

litellm/llms/anthropic/pass_through/messages/mcp_handler.py

@@ -180,7 +180,7 @@ async def anthropic_messages_with_mcp(
         )
 
     if stream:
-        from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
             FakeAnthropicMessagesStreamIterator,
         )
 

litellm/llms/anthropic/pass_through/messages/response_cache.py

@@ -6,7 +6,7 @@ from typing import TYPE_CHECKING, Final
 import litellm
 from litellm._logging import verbose_logger
 from litellm.caching.caching_handler import create_cache_write_task
-from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
     AnthropicMessagesStreamingResponse,
     BaseAnthropicMessagesStreamingIterator,
     _is_message_stop_chunk,

litellm/llms/anthropic/pass_through/messages/streaming_iterator.py

@@ -16,7 +16,7 @@ from litellm.litellm_core_utils.core_helpers import process_response_headers
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
 from litellm.litellm_core_utils.logging_worker import GLOBAL_LOGGING_WORKER
 from litellm.llms.anthropic.common_utils import ANTHROPIC_ERROR_STATUS_CODE_MAP
-from litellm.llms.anthropic.experimental_pass_through.messages.utils import INCOMPLETE_STREAM_ERROR_MESSAGE
+from litellm.llms.anthropic.pass_through.messages.utils import INCOMPLETE_STREAM_ERROR_MESSAGE
 from litellm.proxy.pass_through_endpoints.success_handler import (
     PassThroughEndpointLogging,
 )

litellm/llms/anthropic/pass_through/messages/transformation.py

@@ -590,7 +590,7 @@ class AnthropicMessagesConfig(BaseAnthropicMessagesConfig):
         litellm_logging_obj: LiteLLMLoggingObj,
     ) -> AsyncIterator:
         """Helper function to handle Anthropic streaming responses using the existing logging handlers"""
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             BaseAnthropicMessagesStreamingIterator,
         )
 

litellm/llms/anthropic/pass_through/responses_adapters/handler.py

@@ -109,7 +109,7 @@ def _build_responses_kwargs(
     if isinstance(reasoning, dict):
         effort: Final[object] = reasoning.get("effort")
         if isinstance(effort, str):
-            from litellm.llms.anthropic.experimental_pass_through.utils import (
+            from litellm.llms.anthropic.pass_through.utils import (
                 normalize_reasoning_effort_value,
             )
 

litellm/llms/anthropic/pass_through/responses_adapters/streaming_iterator.py

@@ -15,7 +15,7 @@ from litellm.exceptions import MidStreamFallbackError
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
     encrypted_reasoning_signature,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+from litellm.llms.anthropic.pass_through.messages.utils import (
     INCOMPLETE_STREAM_ERROR_MESSAGE,
     refusal_stop_details,
     responses_output_refusal_text,

litellm/llms/anthropic/pass_through/responses_adapters/transformation.py

@@ -20,11 +20,11 @@ from litellm.litellm_core_utils.prompt_templates.common_utils import (
 from litellm.litellm_core_utils.reasoning_effort_utils import (
     reasoning_effort_from_thinking_budget,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+from litellm.llms.anthropic.pass_through.messages.utils import (
     refusal_stop_details,
     responses_output_refusal_text,
 )
-from litellm.llms.anthropic.experimental_pass_through.utils import (
+from litellm.llms.anthropic.pass_through.utils import (
     is_reasoning_auto_summary_enabled,
     prompt_cache_key_from_user_id,
 )
@@ -69,7 +69,7 @@ class LiteLLMAnthropicToResponsesAPIAdapter:
         if raw_usage is None:
             return AnthropicUsage(input_tokens=0, output_tokens=0)
 
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
         from litellm.responses.utils import ResponseAPILoggingUtils

litellm/llms/anthropic/prompt_cache_prediction.py

@@ -16,7 +16,7 @@ import litellm
 from litellm.llms.anthropic.common_utils import AnthropicModelInfo, is_anthropic_oauth_key
 from litellm.llms.anthropic.count_tokens.handler import AnthropicCountTokensHandler
 from litellm.llms.anthropic.count_tokens.transformation import COUNT_TOKEN_OPTION_NAMES
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     DEFAULT_ANTHROPIC_API_VERSION,
     AnthropicMessagesConfig,
 )

litellm/llms/azure_ai/anthropic/messages_transformation.py

@@ -4,7 +4,7 @@ Azure Anthropic messages transformation config - extends AnthropicMessagesConfig
 
 from typing import Any, Final
 
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.azure.common_utils import BaseAzureLLM

litellm/llms/bedrock/chat/converse_transformation.py

@@ -1831,7 +1831,7 @@ class AmazonConverseConfig(BaseConfig):
         anthropic_beta_list: list,
     ) -> None:
         """Keep only compact_20260112 edits for Bedrock; add beta header or drop field."""
-        from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+        from litellm.llms.anthropic.pass_through.context_management.constants import (
             COMPACT_EDIT_TYPE,
         )
         from litellm.types.llms.anthropic import ANTHROPIC_BETA_HEADER_VALUES

litellm/llms/bedrock/claude_platform/messages_transformation.py

@@ -1,7 +1,7 @@
 from typing import Any, Final
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     DEFAULT_ANTHROPIC_API_VERSION,
     AnthropicMessagesConfig,
 )

litellm/llms/bedrock/messages/invoke_transformations/anthropic_claude3_transformation.py

@@ -18,7 +18,7 @@ from litellm.llms.anthropic.chat.transformation import (
     AnthropicConfig,
 )
 from litellm.llms.anthropic.common_utils import AnthropicModelInfo
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.base_llm.anthropic_messages.transformation import (
@@ -798,7 +798,7 @@ class AmazonAnthropicClaudeMessagesConfig(
         merge them from ``message_start`` so logging/cost sees a consistent usage
         object (fixes negative input costs: LIT-2411).
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             BaseAnthropicMessagesStreamingIterator,
         )
 

litellm/llms/bedrock/messages/mantle_transformation.py

@@ -14,7 +14,7 @@ from typing import TYPE_CHECKING, Any, Final
 import httpx
 from pydantic import TypeAdapter
 
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     DEFAULT_ANTHROPIC_API_VERSION,
     AnthropicMessagesConfig,
 )

litellm/llms/custom_httpx/llm_http_handler.py

@@ -204,7 +204,7 @@ if TYPE_CHECKING:
     from litellm.integrations.custom_logger import CustomLogger
     from litellm.litellm_core_utils.litellm_logging import Logging as _LiteLLMLoggingObj
     from litellm.litellm_core_utils.tokenizer import Encoding as Tokenizer
-    from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
         FakeAnthropicMessagesStreamIterator,
     )
     from litellm.llms.base_llm.passthrough.transformation import BasePassthroughConfig
@@ -2124,7 +2124,7 @@ class BaseLLMHTTPHandler:
 
         initial_response: AsyncIterator | AnthropicMessagesResponse
         if stream:
-            from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+            from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
                 AnthropicMessagesStreamingResponse,
                 anthropic_messages_stream_hidden_params,
             )
@@ -2148,7 +2148,7 @@ class BaseLLMHTTPHandler:
                     hidden_params=stream_hidden_params,
                 )
 
-            from litellm.llms.anthropic.experimental_pass_through.messages.agentic_streaming_iterator import (
+            from litellm.llms.anthropic.pass_through.messages.agentic_streaming_iterator import (
                 AgenticAnthropicStreamingIterator,
             )
 
@@ -5540,7 +5540,7 @@ class BaseLLMHTTPHandler:
             from typing import cast
 
             from litellm._logging import verbose_logger
-            from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+            from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
                 FakeAnthropicMessagesStreamIterator,
             )
             from litellm.types.llms.anthropic_messages.anthropic_response import (

litellm/llms/deepseek/messages/transformation.py

@@ -5,7 +5,7 @@ DeepSeek Anthropic-compatible messages transformation config.
 from typing import Any, Final
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.secret_managers.main import get_secret_str

litellm/llms/github_copilot/messages/transformation.py

@@ -1,7 +1,7 @@
 from typing import Any, Final
 
 from litellm.exceptions import AuthenticationError
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 

litellm/llms/minimax/messages/transformation.py

@@ -5,7 +5,7 @@ MiniMax Anthropic transformation config - extends AnthropicConfig for MiniMax's
 from typing import Any, Final  # noqa: TID251  # override below must mirror the legacy base signature
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.secret_managers.main import get_secret_str

litellm/llms/openai_like/messages/transformation.py

@@ -2,7 +2,7 @@ from typing import Any, Final
 
 import litellm
 from litellm.llms.anthropic.common_utils import normalize_cache_control_in_anthropic_payload
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.openai_like.json_loader import SimpleProviderConfig

litellm/llms/tencent/messages/transformation.py

@@ -8,7 +8,7 @@ alongside its standard OpenAI-compatible chat completions endpoint.
 from typing import Any
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.secret_managers.main import get_secret_str

litellm/llms/vertex_ai/vertex_ai_partner_models/anthropic/experimental_pass_through/transformation.py

@@ -1,7 +1,7 @@
 from typing import Any, Final
 
 from litellm.llms.anthropic.common_utils import AnthropicModelInfo
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.types.llms.anthropic import (

litellm/messages/dispatch.py

@@ -5,7 +5,7 @@ from typing import Final, TypeAlias, cast  # noqa: TID251  # native binding sele
 
 from litellm.exceptions import BadRequestError
 from litellm.litellm_core_utils.get_llm_provider_logic import get_llm_provider
-from litellm.llms.anthropic.experimental_pass_through.messages import handler as main
+from litellm.llms.anthropic.pass_through.messages import handler as main
 from litellm.rust_bridge.catalog import Delivery, Route, RouteContext
 from litellm.rust_bridge.dispatch import PublicDispatch, call_hook
 from litellm.rust_bridge.messages.entrypoints import (

litellm/proxy/anthropic_endpoints/endpoints.py

@@ -14,7 +14,7 @@ from litellm.anthropic_interface.exceptions import (
     AnthropicExceptionMapping,
 )
 from litellm.integrations.custom_guardrail import ModifyResponseException
-from litellm.llms.anthropic.experimental_pass_through.context_management import (
+from litellm.llms.anthropic.pass_through.context_management import (
     AnthropicContextManagementError,
 )
 from litellm.llms.base_llm.guardrail_translation.utils import (

litellm/proxy/guardrails/anthropic_sse.py

@@ -167,10 +167,10 @@ def is_sse_error_stream(all_chunks: Sequence[object]) -> bool:
 
 
 def anthropic_sse_chunks_from_response(assembled: ModelResponse) -> tuple[bytes, ...]:
-    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         LiteLLMAnthropicMessagesAdapter,
     )
-    from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
         FakeAnthropicMessagesStreamIterator,
     )
 

litellm/proxy/guardrails/guardrail_hooks/straiker/straiker.py

@@ -535,7 +535,7 @@ def _v3_answer(request_data: Mapping[str, object], model: str | None) -> Mapping
         return _v3_text_completion_as_chat(response)
     if not isinstance(response, ModelResponse) or not _v3_anthropic_messages_route(request_data):
         return _jsonable_dict(response)
-    from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         LiteLLMAnthropicMessagesAdapter,
     )
 

litellm/proxy/pass_through_endpoints/streaming_handler.py

@@ -294,7 +294,7 @@ class PassThroughStreamingHandler:
         - Vertex AI
         - OpenAI
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             _is_message_stop_chunk,  # pyright: ignore[reportPrivateUsage]  # both native stream paths share terminal-event detection
             _is_provider_error_chunk,  # pyright: ignore[reportPrivateUsage]  # provider errors must not become cache evidence
         )

litellm/router.py

@@ -525,7 +525,7 @@ MAX_BUFFERED_PRE_CONTENT_ANTHROPIC_CHUNKS: Final = 200
 def _anthropic_stream_should_drop_pre_content_ping(chunk: object, has_generated_content: bool) -> bool:
     """A `ping` keepalive seen before any real content is dropped outright - it recurs indefinitely on a
     slow-starting connection and carries nothing worth buffering toward a possible fallback."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import is_anthropic_ping_chunk
+    from litellm.llms.anthropic.pass_through.messages.streaming_iterator import is_anthropic_ping_chunk
 
     if has_generated_content:
         return False
@@ -536,7 +536,7 @@ def _anthropic_stream_forwards_ping_live(chunk: object, has_generated_content: b
     """A `ping` that no lifecycle frame precedes reaches the client live: a fallback's own message_start can still
     follow it without overlapping lifecycles, and AgenticAnthropicStreamingIterator's hold-back keepalive is exactly
     such a ping."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import is_anthropic_ping_chunk
+    from litellm.llms.anthropic.pass_through.messages.streaming_iterator import is_anthropic_ping_chunk
 
     if has_generated_content or buffered_chunk_count:
         return False
@@ -550,7 +550,7 @@ def _is_retriable_anthropic_status(status_code: int) -> bool:
 def _anthropic_stream_error_is_gateway_verdict(chunk: object) -> bool:
     """AgenticAnthropicStreamingIterator's own retrieval-failure frame is the gateway's verdict, not a provider
     failure: another deployment would rerun the same failed hook, so it reaches the client instead of falling back."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.agentic_streaming_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.agentic_streaming_iterator import (
         is_server_fulfilled_tool_leak_error,
     )
 
@@ -605,7 +605,7 @@ def _anthropic_stream_commits_now(chunk: object, has_generated_content: bool, bu
     pre-content buffer cap was hit) rather than keep buffering lifecycle
     frames toward a possible fallback.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
         is_anthropic_content_delta_chunk,
     )
 
@@ -5354,7 +5354,7 @@ class Router:
                 response=response,
                 kwargs=kwargs,
             ):
-                from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+                from litellm.llms.anthropic.pass_through.messages.utils import (
                     safeguard_refusal_error,
                 )
 
@@ -5464,12 +5464,12 @@ class Router:
         anyway) or once the stream ends without ever producing content or
         an error.
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             aclose_if_supported,
             parse_anthropic_error_event,
             parse_anthropic_refusal_stop_details,
         )
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             safeguard_refusal_error,
         )
 
@@ -5626,7 +5626,7 @@ class Router:
         budget.
         """
         from litellm.exceptions import MidStreamFallbackError
-        from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
             aclose_if_supported,
             anthropic_messages_response_as_sse_events,
         )
@@ -8466,7 +8466,7 @@ class Router:
         when a content-policy fallback is configured; a plain refusal without stop_details, or
         any response with nothing configured, is returned to the client unchanged.
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             get_safeguard_refusal_stop_details,
         )
 
@@ -12206,7 +12206,7 @@ class Router:
         `tools` (Chat Completions, Responses and Anthropic Messages shapes) and the
         Anthropic Messages top-level `system` block.
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.utils import (
+        from litellm.llms.anthropic.pass_through.messages.utils import (
             anthropic_system_to_openai_message,
         )
 

litellm/rust_bridge/callbacks_legacy_python.py

@@ -326,7 +326,7 @@ def stream_success(
     end: datetime.datetime,
     first_chunk: datetime.datetime | None,
 ) -> None:
-    from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
         GLOBAL_PASS_THROUGH_SUCCESS_HANDLER_OBJ,
     )
     from litellm.proxy.pass_through_endpoints.streaming_handler import PassThroughStreamingHandler

litellm/rust_bridge/messages/route_host.py

@@ -9,7 +9,7 @@ from pydantic import TypeAdapter, ValidationError
 
 import litellm
 from litellm.litellm_core_utils.core_helpers import normalize_drop_params
-from litellm.llms.anthropic.experimental_pass_through.utils import is_reasoning_auto_summary_enabled
+from litellm.llms.anthropic.pass_through.utils import is_reasoning_auto_summary_enabled
 from litellm.rust_bridge import failures
 from litellm.rust_bridge.messages.entrypoints import LiteLLMMessagesRequest
 from litellm.types.llms.anthropic_messages.anthropic_response import AnthropicMessagesResponse
@@ -55,7 +55,7 @@ def response(value: Mapping[str, object]) -> AnthropicMessagesResponse:
 
 
 def stream_hidden_params(headers: Sequence[tuple[str, str]]) -> Mapping[str, object]:
-    from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
         anthropic_messages_stream_hidden_params,
     )
 

ruff-strict.toml

@@ -63,7 +63,7 @@ max-args = 5
 # directly, each with a `# noqa: TID251  # <reason>`.
 "litellm.responses.main.responses".msg = "Import litellm.responses.dispatch.responses so the call routes through dispatch."
 "litellm.responses.main.aresponses".msg = "Import litellm.responses.dispatch.aresponses so the call routes through dispatch."
-"litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages".msg = "Import litellm.messages.anthropic_messages so the call routes through dispatch."
-"litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages_handler".msg = "Import litellm.messages.anthropic_messages_handler so the call routes through dispatch."
+"litellm.llms.anthropic.pass_through.messages.handler.anthropic_messages".msg = "Import litellm.messages.anthropic_messages so the call routes through dispatch."
+"litellm.llms.anthropic.pass_through.messages.handler.anthropic_messages_handler".msg = "Import litellm.messages.anthropic_messages_handler so the call routes through dispatch."
 "litellm.main.completion".msg = "Import litellm.completion so the call routes through dispatch."
 "litellm.main.acompletion".msg = "Import litellm.acompletion so the call routes through dispatch."

tests/e2e/coverage_registry/quota_management.yaml

@@ -41,7 +41,7 @@
 - {id: quota_management.budget.spend_counter.reseed_matches_db, module: quota_management, tier: P2, behavior: budget, variant: spend_counter, assertions: [reseed_matches_db], exercised_on: [chat_completions], source: "proxy/spend_tracking/budget_reservation.py", rationale: "Concurrent cold-counter reseeds keep the enforcement counter equal to DB spend (#26829)"}
 - {id: quota_management.spend_tracking.chat_completions.logs_cost, module: quota_management, tier: P0, behavior: spend_tracking, variant: chat_completions, assertions: [logs_cost], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "A paid chat call writes a nonzero spend row"}
 - {id: quota_management.spend_tracking.stream.logs_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: stream, assertions: [logs_cost], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "Streaming responses aggregate token counts into a spend row"}
-- {id: quota_management.spend_tracking.messages_bridge.logs_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: messages_bridge, assertions: [logs_cost], exercised_on: [messages], source: "llms/anthropic/experimental_pass_through/responses_adapters/handler.py", rationale: "A streaming /v1/messages request served by an openai-provider model is bridged through the anthropic-messages -> Responses adapter and must aggregate the consumed SSE stream into one spend row with nonzero cost and token counts, attributed to custom_llm_provider openai under call_type anthropic_messages"}
+- {id: quota_management.spend_tracking.messages_bridge.logs_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: messages_bridge, assertions: [logs_cost], exercised_on: [messages], source: "llms/anthropic/pass_through/responses_adapters/handler.py", rationale: "A streaming /v1/messages request served by an openai-provider model is bridged through the anthropic-messages -> Responses adapter and must aggregate the consumed SSE stream into one spend row with nonzero cost and token counts, attributed to custom_llm_provider openai under call_type anthropic_messages"}
 - {id: quota_management.spend_tracking.embeddings.logs_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: embeddings, assertions: [logs_cost], exercised_on: [embeddings], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "Embedding calls write nonzero spend rows"}
 - {id: quota_management.spend_tracking.cache_hit.zero_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: cache_hit, assertions: [zero_cost], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "A response-cache hit logs at zero cost with the cache-hit marker"}
 - {id: quota_management.spend_tracking.key_rollup.matches_sum_of_logs, module: quota_management, tier: P1, behavior: spend_tracking, variant: key_rollup, assertions: [matches_sum_of_logs], exercised_on: [chat_completions], source: "proxy/db/db_spend_update_writer.py", rationale: "A key's rolled-up spend equals the sum of its log rows"}
@@ -59,7 +59,7 @@
 - {id: quota_management.spend_tracking.cache_write.bills_cache_creation_rate, module: quota_management, tier: P1, behavior: spend_tracking, variant: cache_write, assertions: [bills_cache_creation_rate], exercised_on: [chat_completions], source: "litellm_core_utils/llm_cost_calc/utils.py", rationale: "OpenAI cache-write tokens land on the spend row as cache-creation tokens billed at the cache-creation rate, not silently at the input rate (#34046)"}
 - {id: quota_management.spend_tracking.cost_breakdown.reports_component_costs, module: quota_management, tier: P1, behavior: spend_tracking, variant: cost_breakdown, assertions: [reports_component_costs], exercised_on: [chat_completions], source: "proxy/spend_tracking/spend_tracking_utils.py", rationale: "The spend row's metadata.cost_breakdown itemizes cache-read, cache-creation, output, and reasoning costs at the deployment's own rates and they sum to the row's spend (#31686)"}
 - {id: quota_management.spend_tracking.stream_cache_read.bills_cache_read_rate, module: quota_management, tier: P1, behavior: spend_tracking, variant: stream_cache_read, assertions: [bills_cache_read_rate], exercised_on: [chat_completions], source: "litellm_core_utils/streaming_chunk_builder_utils.py", rationale: "A streamed call's reassembled usage keeps the cached-token detail so cache reads bill at the cache-read discount, not full input price (#34812)"}
-- {id: quota_management.spend_tracking.messages_bridge.keeps_cache_tokens, module: quota_management, tier: P1, behavior: spend_tracking, variant: messages_bridge, assertions: [keeps_cache_tokens], exercised_on: [messages], source: "llms/anthropic/experimental_pass_through/responses_adapters/handler.py", rationale: "A /v1/messages request served by a Responses-only OpenAI model keeps its cache-read tokens and their discounted billing across the bridge (#34957)"}
+- {id: quota_management.spend_tracking.messages_bridge.keeps_cache_tokens, module: quota_management, tier: P1, behavior: spend_tracking, variant: messages_bridge, assertions: [keeps_cache_tokens], exercised_on: [messages], source: "llms/anthropic/pass_through/responses_adapters/handler.py", rationale: "A /v1/messages request served by a Responses-only OpenAI model keeps its cache-read tokens and their discounted billing across the bridge (#34957)"}
 - {id: quota_management.spend_tracking.service_tier.bills_tier_rates, module: quota_management, tier: P1, behavior: spend_tracking, variant: service_tier, assertions: [bills_tier_rates], exercised_on: [chat_completions], source: "cost_calculator.py", rationale: "A priority service_tier call bills input, output, and reasoning at the deployment's *_priority rates and records the tier on the row (#35923, #35925)"}
 - {id: quota_management.spend_tracking.cost_headers.additive_components, module: quota_management, tier: P1, behavior: spend_tracking, variant: cost_headers, assertions: [additive_components], exercised_on: [chat_completions], source: "proxy/common_request_processing.py", rationale: "The x-litellm-response-cost-* component headers sum to the total, input covers only fresh tokens, and reasoning stays a subset of output (#36965)"}
 - {id: quota_management.spend_tracking.passthrough_stream.injects_usage_cost, module: quota_management, tier: P1, behavior: spend_tracking, variant: passthrough_stream, assertions: [injects_usage_cost], exercised_on: [openai_passthrough], source: "proxy/pass_through_endpoints/streaming_handler.py", rationale: "With include_cost_in_streaming_usage on, the /openai passthrough's final streaming usage frame carries the proxy-computed cost (#36503). Uncovered: the flag is only settable in litellm_settings, and the shared e2e stack does not turn it on yet"}

tests/pass_through_unit_tests/base_anthropic_unified_messages_test.py

@@ -8,7 +8,7 @@ from unittest.mock import AsyncMock, MagicMock
 import litellm
 import pytest
 from dotenv import load_dotenv
-from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+from litellm.llms.anthropic.pass_through.messages.handler import (
     anthropic_messages,
 )
 

tests/pass_through_unit_tests/test_anthropic_messages_passthrough.py

@@ -9,7 +9,7 @@ from unittest.mock import AsyncMock, MagicMock
 import litellm
 import pytest
 from dotenv import load_dotenv
-from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+from litellm.llms.anthropic.pass_through.messages.handler import (
     anthropic_messages,
 )
 

tests/pass_through_unit_tests/test_context_management_polyfill.py

@@ -6,7 +6,7 @@ from unittest.mock import patch
 import pytest
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+from litellm.llms.anthropic.pass_through.context_management.constants import (
     CLEARED_TOOL_RESULT_PLACEHOLDER,
 )
 from litellm.types.utils import (

tests/pass_through_unit_tests/test_websearch_interception_e2e.py

@@ -993,7 +993,7 @@ async def test_pre_request_hook_modifies_request_body():
 
     # Patch the anthropic_messages_handler function (called after hooks)
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.handler.anthropic_messages_handler",
         side_effect=mock_anthropic_messages_handler,
     ), patch(  # test-quality-ok: the hook imports this process-global router at call time; no injection seam exists to register search_tools
         "litellm.proxy.proxy_server.llm_router",

tests/test_litellm/proxy/guardrails/guardrail_hooks/test_headroom.py

@@ -2650,7 +2650,7 @@ async def test_retrieved_content_protected_when_mcp_tool_name_is_truncated(guard
     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 (
+    from litellm.llms.anthropic.pass_through.adapters.transformation import (
         truncate_tool_name,
     )
 

tests/test_litellm/proxy/spend_tracking/test_spend_tracking_utils.py

@@ -4089,7 +4089,7 @@ async def test_spend_log_request_id_is_the_message_id_a_bridged_streaming_caller
     adapter mints itself, and it is the only request id that call ever shows the caller, so
     GET /spend/logs?request_id=msg_... has to land on the row."""
     from litellm.litellm_core_utils.litellm_logging import Logging
-    from litellm.llms.anthropic.experimental_pass_through.responses_adapters.streaming_iterator import (
+    from litellm.llms.anthropic.pass_through.responses_adapters.streaming_iterator import (
         AnthropicResponsesStreamWrapper,
     )
     from litellm.types.llms.openai import (

tests/test_litellm/proxy/test_budget_reservation.py

@@ -13,10 +13,10 @@ import litellm
 from litellm.caching.dual_cache import DualCache
 from litellm.types.caching import RedisPipelineIncrementOperation
 from litellm.constants import STREAM_SSE_KEEPALIVE_PING_BYTES
-from litellm.llms.anthropic.experimental_pass_through.messages.agentic_streaming_iterator import (
+from litellm.llms.anthropic.pass_through.messages.agentic_streaming_iterator import (
     AgenticAnthropicStreamingIterator,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
     AnthropicMessagesStreamingResponse,
 )
 from litellm.proxy._types import (

tests/test_litellm/proxy/utils/proxy_logging/test_streaming_hooks.py

@@ -23,7 +23,7 @@ from litellm.exceptions import GuardrailRaisedException
 from litellm.integrations.custom_guardrail import CustomGuardrail
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.litellm_core_utils.litellm_logging import Logging as LiteLLMLoggingObj
-from litellm.llms.anthropic.experimental_pass_through.messages.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.messages.streaming_iterator import (
     BaseAnthropicMessagesStreamingIterator,
 )
 from litellm.proxy._types import UserAPIKeyAuth

tests/unit/integrations/websearch_interception/test_websearch_agentic_loop_cap.py

@@ -25,7 +25,7 @@ from litellm.integrations.websearch_interception.handler import (
 )
 from litellm.integrations.websearch_interception.tools import get_litellm_web_search_tool
 from litellm.litellm_core_utils.agentic_loop_settings import DEFAULT_MAX_AGENTIC_LOOPS
-from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
     FakeAnthropicMessagesStreamIterator,
 )
 from litellm.llms.base_llm.search.transformation import SearchResponse, SearchResult

tests/unit/integrations/websearch_interception/test_websearch_short_circuit.py

@@ -229,7 +229,7 @@ class TestShortCircuitEntryPoint:
     @pytest.mark.asyncio
     async def test_returns_none_when_no_callbacks(self):
         """No callbacks configured → returns None"""
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 
@@ -246,7 +246,7 @@ class TestShortCircuitEntryPoint:
     @pytest.mark.asyncio
     async def test_returns_dict_when_not_streaming(self):
         """Non-streaming short-circuit → returns dict"""
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 
@@ -271,10 +271,10 @@ class TestShortCircuitEntryPoint:
     @pytest.mark.asyncio
     async def test_returns_stream_iterator_when_streaming(self):
         """Streaming short-circuit → returns FakeAnthropicMessagesStreamIterator"""
-        from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
             FakeAnthropicMessagesStreamIterator,
         )
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 
@@ -313,7 +313,7 @@ class TestShortCircuitEntryPoint:
         """Non-WebSearchInterceptionLogger callbacks are ignored"""
         from unittest.mock import MagicMock
 
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 
@@ -336,10 +336,10 @@ class TestShortCircuitEntryPoint:
         loop. The short-circuit must use the ORIGINAL stream value so streaming
         callers get SSE events instead of a plain dict.
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+        from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
             FakeAnthropicMessagesStreamIterator,
         )
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 
@@ -369,7 +369,7 @@ class TestShortCircuitEntryPoint:
         still fire the short-circuit when the caller propagates the derived
         provider.
         """
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             _try_websearch_short_circuit,
         )
 

tests/unit/integrations/websearch_interception/test_websearch_streaming_wrap.py

@@ -13,7 +13,7 @@ import pytest
 
 from litellm.integrations.custom_logger import CustomLogger
 from litellm.llms.custom_httpx.llm_http_handler import BaseLLMHTTPHandler
-from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
     FakeAnthropicMessagesStreamIterator,
 )
 from litellm.types.integrations.custom_logger import AgenticLoopPlan

tests/unit/llms/anthropic/chat/test_anthropic_chat_transformation.py

@@ -19,7 +19,7 @@ from litellm.constants import (
 )
 from litellm.litellm_core_utils.prompt_templates.common_utils import encrypted_reasoning_signature
 from litellm.llms.anthropic.chat.transformation import AnthropicConfig
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.azure_ai.anthropic.transformation import AzureAnthropicConfig
@@ -5783,7 +5783,7 @@ def test_translate_system_message_strips_billing_header_for_bedrock_invoke():
     [
         ("litellm.llms.anthropic.chat.transformation", "AnthropicConfig", False),
         (
-            "litellm.llms.anthropic.experimental_pass_through.messages.transformation",
+            "litellm.llms.anthropic.pass_through.messages.transformation",
             "AnthropicMessagesConfig",
             False,
         ),

tests/unit/llms/anthropic/messages/test_advisor_orchestration.py

@@ -68,7 +68,7 @@ def _make_advisor_tool_use_response(
 
 
 def test_can_handle_edge_cases():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -96,7 +96,7 @@ async def test_anthropic_native_interceptor_skipped():
     For provider=anthropic, can_handle() must return False.
     The interceptor must never call handle().
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -114,7 +114,7 @@ async def test_anthropic_native_interceptor_skipped():
 @pytest.mark.asyncio
 async def test_loop_no_advisor_call():
     """Executor returns text on first try — no advisor call, loop exits immediately."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
         _call_messages_handler,
     )
@@ -123,7 +123,7 @@ async def test_loop_no_advisor_call():
     executor_response = _make_text_response(final_text)
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         new_callable=AsyncMock,
         return_value=executor_response,
     ) as mock_call:
@@ -156,7 +156,7 @@ async def test_loop_one_advisor_call():
     Executor calls advisor once → advisor responds → executor produces final text.
     Total calls: 3 (executor, advisor, executor-final).
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -184,7 +184,7 @@ async def test_loop_one_advisor_call():
         return final_resp  # executor: final answer
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -218,7 +218,7 @@ async def test_loop_one_advisor_call():
 @pytest.mark.asyncio
 async def test_loop_max_uses_raises():
     """Loop exceeding max_uses must raise AdvisorMaxIterationsError."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorMaxIterationsError,
         AdvisorOrchestrationHandler,
     )
@@ -239,7 +239,7 @@ async def test_loop_max_uses_raises():
         return advisor_tool_use_resp
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -262,17 +262,17 @@ async def test_loop_max_uses_raises():
 @pytest.mark.asyncio
 async def test_loop_streaming_wraps_response():
     """stream=True: final response must be wrapped in FakeAnthropicMessagesStreamIterator."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.fake_stream_iterator import (
+    from litellm.llms.anthropic.pass_through.messages.fake_stream_iterator import (
         FakeAnthropicMessagesStreamIterator,
     )
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
     executor_response = _make_text_response("Hello, world!")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         new_callable=AsyncMock,
         return_value=executor_response,
     ):
@@ -308,7 +308,7 @@ async def test_prior_advisor_blocks_replaced_in_history():
     History containing server_tool_use + advisor_tool_result blocks gets
     collapsed to <advisor_feedback> text before forwarding to the executor.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -341,7 +341,7 @@ async def test_prior_advisor_blocks_replaced_in_history():
         return _make_text_response("Here is the efficient version.")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -383,7 +383,7 @@ async def test_advisor_tool_translated_for_executor():
     """
     The executor must receive a regular tool definition (not advisor_20260301 type).
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -395,7 +395,7 @@ async def test_advisor_tool_translated_for_executor():
         return _make_text_response("Done.")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -425,7 +425,7 @@ async def test_advisor_tool_translated_for_executor():
 @pytest.mark.asyncio
 async def test_max_uses_zero_raises_on_first_advisor_call():
     """max_uses=0 must cause AdvisorMaxIterationsError on the first advisor call."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorMaxIterationsError,
         AdvisorOrchestrationHandler,
     )
@@ -437,7 +437,7 @@ async def test_max_uses_zero_raises_on_first_advisor_call():
         return advisor_tool_use_resp  # executor always tries to call advisor
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -460,7 +460,7 @@ async def test_max_uses_zero_raises_on_first_advisor_call():
 @pytest.mark.asyncio
 async def test_missing_advisor_model_raises_value_error():
     """handle() must raise ValueError when the advisor tool has no model field."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -487,7 +487,7 @@ async def test_missing_advisor_model_raises_value_error():
 async def test_max_uses_none_falls_back_to_default():
     """When max_uses is absent, the handler uses ADVISOR_MAX_USES from constants."""
     import litellm.constants as _c
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorMaxIterationsError,
         AdvisorOrchestrationHandler,
     )
@@ -501,7 +501,7 @@ async def test_max_uses_none_falls_back_to_default():
         return advisor_tool_use_resp
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -535,7 +535,7 @@ ADVISOR_TOOL_WITH_CREDS = {
 
 async def _run_advisor_and_capture_subcall_kwargs():
     """Run one advisor turn and return the kwargs of the advisor sub-call."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -560,11 +560,11 @@ async def _run_advisor_and_capture_subcall_kwargs():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_call,
         ),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor.validate_url",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor.validate_url",
         ),
     ):
         h = AdvisorOrchestrationHandler()
@@ -584,7 +584,7 @@ async def test_advisor_creds_dropped_when_proxy_opt_in_disabled():
     """On the proxy without opt-in, the caller's advisor api_base/api_key must
     NOT reach the sub-call (would redirect it / leak the server key)."""
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=False,
     ):
         captured = await _run_advisor_and_capture_subcall_kwargs()
@@ -596,7 +596,7 @@ async def test_advisor_creds_dropped_when_proxy_opt_in_disabled():
 async def test_advisor_creds_honored_when_proxy_opt_in_enabled():
     """With the admin opt-in, the documented clientside routing still works."""
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=True,
     ):
         captured = await _run_advisor_and_capture_subcall_kwargs()
@@ -629,7 +629,7 @@ def test_allow_client_side_advisor_credentials_reads_proxy_flag():
     """The gate mirrors the proxy's allow_client_side_credentials opt-in."""
     import sys
 
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _allow_client_side_advisor_credentials,
     )
 
@@ -653,7 +653,7 @@ def test_allow_client_side_advisor_credentials_defaults_true_outside_proxy():
     import builtins
     import sys
 
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _allow_client_side_advisor_credentials,
     )
 
@@ -677,7 +677,7 @@ def test_advisor_gate_propagates_non_import_errors():
     returning True."""
     import sys
 
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors import (
         advisor,
     )
 
@@ -744,12 +744,12 @@ async def test_advisor_uses_tool_credentials_when_clientside_enabled():
 
 
 def test_resolve_advisor_credentials_returns_none_when_gate_closed():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=False,
     ):
         result = _resolve_advisor_credentials(ADVISOR_TOOL_WITH_CREDS)
@@ -757,18 +757,18 @@ def test_resolve_advisor_credentials_returns_none_when_gate_closed():
 
 
 def test_resolve_advisor_credentials_allows_api_key_without_api_base():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     tool = {**ADVISOR_TOOL, "api_key": "sk-other"}
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
             return_value=True,
         ),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor.validate_url",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor.validate_url",
             side_effect=AssertionError("validate_url must not run without an api_base"),
         ),
     ):
@@ -777,13 +777,13 @@ def test_resolve_advisor_credentials_allows_api_key_without_api_base():
 
 
 def test_resolve_advisor_credentials_rejects_api_base_without_api_key():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     tool = {**ADVISOR_TOOL, "api_base": "https://other.example"}
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=True,
     ):
         with pytest.raises(ValueError, match="api_base"):
@@ -791,17 +791,17 @@ def test_resolve_advisor_credentials_rejects_api_base_without_api_key():
 
 
 def test_resolve_advisor_credentials_validates_api_base_before_use():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
             return_value=True,
         ),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor.validate_url"
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor.validate_url"
         ) as mock_validate,
     ):
         result = _resolve_advisor_credentials(ADVISOR_TOOL_WITH_CREDS)
@@ -811,17 +811,17 @@ def test_resolve_advisor_credentials_validates_api_base_before_use():
 
 def test_resolve_advisor_credentials_propagates_ssrf_error():
     from litellm.litellm_core_utils.url_utils import SSRFError
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
             return_value=True,
         ),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor.validate_url",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor.validate_url",
             side_effect=SSRFError("URL targets a blocked address"),
         ),
     ):
@@ -832,18 +832,18 @@ def test_resolve_advisor_credentials_propagates_ssrf_error():
 def test_resolve_advisor_credentials_skips_validation_when_url_validation_disabled():
     import litellm
 
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
             return_value=True,
         ),
         patch.object(litellm, "user_url_validation", False),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor.validate_url",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor.validate_url",
             side_effect=AssertionError("validate_url must not run when user_url_validation is disabled"),
         ),
     ):
@@ -855,7 +855,7 @@ def test_resolve_advisor_credentials_blocks_real_cloud_metadata_address():
     """End-to-end (no mocked validate_url): a caller can't redirect the
     advisor sub-call to the cloud-metadata address even with an api_key."""
     from litellm.litellm_core_utils.url_utils import SSRFError
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
@@ -865,7 +865,7 @@ def test_resolve_advisor_credentials_blocks_real_cloud_metadata_address():
         "api_base": "https://169.254.169.254/latest/meta-data/",
     }
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=True,
     ):
         with pytest.raises(SSRFError):
@@ -873,13 +873,13 @@ def test_resolve_advisor_credentials_blocks_real_cloud_metadata_address():
 
 
 def test_resolve_advisor_credentials_rejects_non_https_api_base():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     tool = {**ADVISOR_TOOL, "api_key": "sk-other", "api_base": "http://8.8.8.8"}
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=True,
     ):
         with pytest.raises(ValueError, match="https"):
@@ -889,14 +889,14 @@ def test_resolve_advisor_credentials_rejects_non_https_api_base():
 def test_resolve_advisor_credentials_rejects_api_base_when_ssl_verify_disabled():
     import litellm
 
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     tool = {**ADVISOR_TOOL, "api_key": "sk-other", "api_base": "https://8.8.8.8"}
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
             return_value=True,
         ),
         patch.object(litellm, "ssl_verify", False),
@@ -908,13 +908,13 @@ def test_resolve_advisor_credentials_rejects_api_base_when_ssl_verify_disabled()
 def test_resolve_advisor_credentials_allows_real_public_ip_address():
     """End-to-end (no mocked validate_url): a globally-routable literal IP
     api_base is honored when paired with an api_key."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         _resolve_advisor_credentials,
     )
 
     tool = {**ADVISOR_TOOL, "api_key": "sk-other", "api_base": "https://8.8.8.8"}
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._allow_client_side_advisor_credentials",
         return_value=True,
     ):
         result = _resolve_advisor_credentials(tool)
@@ -933,7 +933,7 @@ async def test_advisor_sub_call_failure_is_tagged():
     """When the advisor sub-call raises, the exception that propagates out of
     handle() must be tagged as an advisor orchestration failure."""
     import litellm
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
     from litellm.router_utils.cooldown_handlers import is_advisor_orchestration_failure
@@ -952,7 +952,7 @@ async def test_advisor_sub_call_failure_is_tagged():
         )
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -975,7 +975,7 @@ async def test_advisor_max_iterations_failure_is_tagged():
     """When the orchestration loop exceeds max_uses (the executor keeps calling
     the advisor), the AdvisorMaxIterationsError must be tagged so the healthy
     executor deployment is not cooled down."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorMaxIterationsError,
         AdvisorOrchestrationHandler,
     )
@@ -991,7 +991,7 @@ async def test_advisor_max_iterations_failure_is_tagged():
         return _make_advisor_tool_use_response()
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -1013,7 +1013,7 @@ async def test_executor_failure_is_not_tagged():
     """A failure of the executor call (not advisor orchestration) must NOT be
     tagged — the selected deployment genuinely failed and should cool down."""
     import litellm
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
     from litellm.router_utils.cooldown_handlers import is_advisor_orchestration_failure
@@ -1026,7 +1026,7 @@ async def test_executor_failure_is_not_tagged():
         )
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()
@@ -1082,7 +1082,7 @@ def _router_with_advisor_deployment(
 @pytest.mark.asyncio
 async def test_advisor_sub_call_routes_through_proxy_router():
     import litellm.proxy.proxy_server as proxy_server
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -1105,7 +1105,7 @@ async def test_advisor_sub_call_routes_through_proxy_router():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_call,
         ),
         patch.object(proxy_server, "llm_router", router),
@@ -1143,7 +1143,7 @@ async def test_advisor_sub_call_routes_through_proxy_router():
 async def test_advisor_sub_call_routes_through_router_for_alias_and_wildcard(router_kwargs, advisor_model):
     """Alias and wildcard advisor models resolve through the router like exact model_list matches."""
     import litellm.proxy.proxy_server as proxy_server
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -1166,7 +1166,7 @@ async def test_advisor_sub_call_routes_through_router_for_alias_and_wildcard(rou
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_call,
         ),
         patch.object(proxy_server, "llm_router", router),
@@ -1192,7 +1192,7 @@ async def test_advisor_sub_call_routes_through_router_for_alias_and_wildcard(rou
 async def test_advisor_sub_call_bypasses_router_for_unconfigured_model():
     """An advisor model the router doesn't know about keeps the SDK-level path."""
     import litellm.proxy.proxy_server as proxy_server
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -1217,7 +1217,7 @@ async def test_advisor_sub_call_bypasses_router_for_unconfigured_model():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_call,
         ),
         patch.object(proxy_server, "llm_router", router),
@@ -1241,7 +1241,7 @@ async def test_advisor_sub_call_client_override_bypasses_router():
     """A caller-supplied api_key/api_base override must not be re-routed."""
     import litellm
     import litellm.proxy.proxy_server as proxy_server
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -1272,7 +1272,7 @@ async def test_advisor_sub_call_client_override_bypasses_router():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_call,
         ),
         patch.object(proxy_server, "llm_router", router),
@@ -1307,7 +1307,7 @@ async def test_advisor_sub_call_client_override_bypasses_router():
 
 @pytest.mark.asyncio
 async def test_advisor_context_excludes_in_sequence_system_rows():
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorOrchestrationHandler,
     )
 
@@ -1327,7 +1327,7 @@ async def test_advisor_context_excludes_in_sequence_system_rows():
         return _make_text_response("Final answer.")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_call,
     ):
         h = AdvisorOrchestrationHandler()

tests/unit/llms/anthropic/pass_through/adapters/test_anthropic_experimental_pass_through_adapters_transformation.py

@@ -16,14 +16,14 @@ from litellm.litellm_core_utils.prompt_templates.factory import (
     THOUGHT_SIGNATURE_SEPARATOR,
     _bedrock_converse_messages_pt,
 )
-from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+from litellm.llms.anthropic.pass_through.adapters.transformation import (
     OPENAI_MAX_TOOL_NAME_LENGTH,
     AnthropicAdapter,
     LiteLLMAnthropicMessagesAdapter,
     create_tool_name_mapping,
     truncate_tool_name,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.mid_conversation_system import (
+from litellm.llms.anthropic.pass_through.messages.mid_conversation_system import (
     CONVERTED_SYSTEM_NOTE,
 )
 from litellm.llms.openai.chat.gpt_transformation import OpenAIGPTConfig
@@ -3944,7 +3944,7 @@ class TestAnthropicStreamWrapperToolArgs:
         return [text_chunk, tool_chunk, finish_chunk]
 
     def _make_stream_wrapper(self, chunks):
-        from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+        from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
             AnthropicStreamWrapper,
         )
 
@@ -4059,7 +4059,7 @@ def _make_simple_openai_response(
 
 def test_translate_openai_response_to_anthropic_with_polyfill_compaction_block():
     """compaction_block from PolyfillResult must be prepended to content at index 0."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.result import (
+    from litellm.llms.anthropic.pass_through.context_management.result import (
         PolyfillResult,
     )
 
@@ -4092,7 +4092,7 @@ def test_translate_openai_response_to_anthropic_with_polyfill_compaction_block()
 
 def test_translate_openai_response_to_anthropic_with_polyfill_iterations_usage():
     """iterations_usage from PolyfillResult must produce usage['iterations'] with a message entry."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.result import (
+    from litellm.llms.anthropic.pass_through.context_management.result import (
         PolyfillResult,
     )
 
@@ -4147,7 +4147,7 @@ def test_translate_openai_response_to_anthropic_no_polyfill_no_change():
 
 def test_translate_openai_response_to_anthropic_with_polyfill_both_compaction_and_iterations():
     """Full summary path: compaction_block and iterations_usage both present simultaneously."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.result import (
+    from litellm.llms.anthropic.pass_through.context_management.result import (
         PolyfillResult,
     )
 

tests/unit/llms/anthropic/pass_through/adapters/test_handler_output_config_passthrough.py

@@ -38,7 +38,7 @@ sys.path.insert(
     0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../../.."))
 )
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+from litellm.llms.anthropic.pass_through.adapters.handler import (
     ANTHROPIC_ONLY_REQUEST_KEYS,
     LiteLLMMessagesToCompletionTransformationHandler,
 )

tests/unit/llms/anthropic/pass_through/adapters/test_handler_prompt_cache_key.py

@@ -6,7 +6,7 @@ import pytest
 
 sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "../../../../../..")))
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+from litellm.llms.anthropic.pass_through.adapters.handler import (
     LiteLLMMessagesToCompletionTransformationHandler,
 )
 

tests/unit/llms/anthropic/pass_through/adapters/test_handler_reasoning_effort_normalization.py

@@ -10,7 +10,7 @@ from typing import Final
 
 import pytest
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+from litellm.llms.anthropic.pass_through.adapters.handler import (
     LiteLLMMessagesToCompletionTransformationHandler,
 )
 

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_combined_chunk.py

@@ -14,7 +14,7 @@ import json
 from types import SimpleNamespace
 from typing import AsyncIterator
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
     _CombinedChunkSplitter,
 )

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_compaction.py

@@ -6,7 +6,7 @@ from unittest.mock import MagicMock
 import pytest
 
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 from litellm.types.utils import Delta, StreamingChoices, Usage

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_empty_choices.py

@@ -12,7 +12,7 @@ import asyncio
 import json
 from typing import Any, AsyncIterator, Dict, List, Optional
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 from litellm.types.utils import Delta, ModelResponseStream, StreamingChoices, Usage

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_first_delta.py

@@ -27,7 +27,7 @@ from unittest.mock import MagicMock
 import pytest
 
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 from litellm.types.utils import (

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_message_id.py

@@ -12,10 +12,10 @@ import pytest
 import respx
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+from litellm.llms.anthropic.pass_through.adapters.handler import (
     LiteLLMMessagesToCompletionTransformationHandler,
 )
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_mid_stream_error.py

@@ -25,7 +25,7 @@ import pytest
 sys.path.insert(0, os.path.abspath("../../../../.."))
 
 from litellm.exceptions import MidStreamFallbackError
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
     _mid_stream_error_sse_event,
 )

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_stop_reason.py

@@ -12,7 +12,7 @@ bridge emitted ``stop_reason: "end_turn"`` and Anthropic tool-runners
 
 import pytest
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 from litellm.llms.ollama.chat.transformation import (

tests/unit/llms/anthropic/pass_through/adapters/test_streaming_iterator_tool_args.py

@@ -16,7 +16,7 @@ from unittest.mock import MagicMock
 import pytest
 
 
-from litellm.llms.anthropic.experimental_pass_through.adapters.streaming_iterator import (
+from litellm.llms.anthropic.pass_through.adapters.streaming_iterator import (
     AnthropicStreamWrapper,
 )
 from litellm.types.utils import (

tests/unit/llms/anthropic/pass_through/context_management/test_clear_tool_uses.py

@@ -4,10 +4,10 @@ Unit tests for the in-gateway `clear_tool_uses_20250919` polyfill editor.
 
 from copy import deepcopy
 
-from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+from litellm.llms.anthropic.pass_through.context_management.constants import (
     CLEARED_TOOL_RESULT_PLACEHOLDER,
 )
-from litellm.llms.anthropic.experimental_pass_through.context_management.editors.clear_tool_uses import (
+from litellm.llms.anthropic.pass_through.context_management.editors.clear_tool_uses import (
     apply_clear_tool_uses_20250919,
 )
 

tests/unit/llms/anthropic/pass_through/context_management/test_compact.py

@@ -21,11 +21,11 @@ import pytest
 from fastapi import HTTPException
 
 import litellm
-from litellm.llms.anthropic.experimental_pass_through.context_management import (
+from litellm.llms.anthropic.pass_through.context_management import (
     AnthropicContextManagementError,
     apply_context_management,
 )
-from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
     _augment_system_with_summary,
     _extract_summary_text,
     _select_last_user_question,
@@ -34,7 +34,7 @@ from litellm.llms.anthropic.experimental_pass_through.context_management.editors
     apply_client_compaction_block_history,
     apply_compact_20260112,
 )
-from litellm.llms.anthropic.experimental_pass_through.context_management.result import (
+from litellm.llms.anthropic.pass_through.context_management.result import (
     PolyfillResult,
 )
 from litellm.proxy.hooks.parallel_request_limiter_v3 import RateLimitUnverifiableError
@@ -315,7 +315,7 @@ async def test_trigger_below_minimum_raises():
 async def test_trigger_at_minimum_does_not_raise():
     """Exactly 50 000 is allowed — only strictly less than 50k is rejected."""
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_compact_20260112(
@@ -340,7 +340,7 @@ async def test_trigger_at_minimum_does_not_raise():
 async def test_opt_in_gating_no_summary_model_configured():
     messages = _simple_messages()
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_compact_20260112(
@@ -367,7 +367,7 @@ async def test_opt_in_gating_no_summary_model_keeps_post_compaction_tail():
     messages = _messages_with_compaction("prior summary text")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_compact_20260112(
@@ -451,7 +451,7 @@ async def test_slice_only_path_with_existing_compaction_block():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=500),  # well under threshold
@@ -488,7 +488,7 @@ async def test_slice_only_no_compaction_block_under_threshold():
     messages = _simple_messages()
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=500),
@@ -521,12 +521,12 @@ async def test_full_summary_path():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),  # over 150k threshold
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             return_value=mock_response,
         ),
@@ -575,7 +575,7 @@ async def test_full_summary_path_uses_router_when_available():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="my-summary-model",
         ),
         patch("litellm.token_counter", return_value=200_000),
@@ -611,12 +611,12 @@ async def test_litellm_metadata_propagated_to_summary_call():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             return_value=mock_response,
         ) as mock_call,
@@ -648,12 +648,12 @@ async def test_summary_call_failed():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             side_effect=RuntimeError("network error"),
         ),
@@ -684,12 +684,12 @@ async def test_summary_extraction_failed_no_tags():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             return_value=mock_response,
         ),
@@ -716,7 +716,7 @@ async def test_pause_after_compaction_ignored_warning():
     """pause_after_compaction: true → warning recorded, request proceeds normally."""
     messages = _simple_messages()
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_compact_20260112(
@@ -739,7 +739,7 @@ async def test_pause_after_compaction_ignored_warning():
 async def test_unsupported_trigger_type_falls_back_to_default():
     messages = _simple_messages()
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_compact_20260112(
@@ -777,12 +777,12 @@ async def test_custom_instructions_used_verbatim():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -822,12 +822,12 @@ async def test_default_instructions_appended_with_no_tool_suffix_when_no_tools()
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -858,12 +858,12 @@ async def test_default_instructions_with_tools_appends_no_tool_suffix():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -892,12 +892,12 @@ async def test_system_prompt_forwarded_to_summary_call_as_string():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -932,12 +932,12 @@ async def test_system_prompt_forwarded_to_summary_call_as_content_blocks():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -975,12 +975,12 @@ async def test_summary_call_carries_prior_compaction_summary_into_system():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -1012,12 +1012,12 @@ async def test_summary_call_omits_system_message_when_system_is_none():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -1053,12 +1053,12 @@ async def test_summary_call_does_not_emit_consecutive_user_turns():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             side_effect=_fake_call_summary_model,
         ),
     ):
@@ -1085,10 +1085,10 @@ async def test_summary_call_sends_default_max_tokens():
     (which require it) don't reject the request and silently fall back to
     ``summary_call_failed``.
     """
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_SUMMARY_MAX_TOKENS,
     )
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -1112,7 +1112,7 @@ async def test_summary_call_sends_default_max_tokens():
 async def test_summary_call_honors_max_tokens_override():
     """Operators can override the default summary ``max_tokens`` via
     ``general_settings.context_management_summary_max_tokens``."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _read_summary_max_tokens_setting,
     )
 
@@ -1129,7 +1129,7 @@ async def test_summary_call_honors_max_tokens_override():
     ):
         assert _read_summary_max_tokens_setting() == 8192
 
-        from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+        from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
             _call_summary_model,
         )
 
@@ -1148,10 +1148,10 @@ def test_summary_max_tokens_setting_falls_back_for_invalid_values():
     """Invalid override values (non-int, non-positive, missing) fall back to
     the compiled default so a typo in ``general_settings`` doesn't break the
     summary call."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_SUMMARY_MAX_TOKENS,
     )
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _read_summary_max_tokens_setting,
     )
 
@@ -1168,10 +1168,10 @@ def test_summary_max_tokens_setting_falls_back_for_invalid_values():
 async def test_summary_call_sends_default_timeout():
     """``timeout`` is set on the summary call so a slow or unresponsive summary
     model cannot hang the parent ``/v1/messages`` request indefinitely."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_SUMMARY_TIMEOUT_SECONDS,
     )
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -1240,12 +1240,12 @@ async def test_summary_model_denied_when_key_not_in_allowlist():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
     ):
@@ -1272,12 +1272,12 @@ async def test_summary_model_denied_when_team_not_in_allowlist():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
     ):
@@ -1303,12 +1303,12 @@ async def test_summary_model_allowed_when_in_key_allowlist():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
     ):
@@ -1336,12 +1336,12 @@ async def test_summary_model_allowed_when_no_user_api_key_auth():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
     ):
@@ -1373,12 +1373,12 @@ async def test_summary_model_denied_when_user_scope_excludes_it():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch(
@@ -1423,12 +1423,12 @@ async def test_summary_model_denied_when_project_scope_excludes_it():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch(
@@ -1475,12 +1475,12 @@ async def test_summary_model_denied_when_team_member_scope_excludes_it():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch(
@@ -1528,12 +1528,12 @@ async def test_summary_model_denied_when_team_membership_read_hits_a_db_outage()
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch(
@@ -1582,12 +1582,12 @@ async def test_summary_model_denied_when_key_over_model_budget():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.model_max_budget_limiter", limiter),
@@ -1635,12 +1635,12 @@ async def test_summary_model_denied_when_user_over_model_budget():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.model_max_budget_limiter", limiter),
@@ -1702,12 +1702,12 @@ async def test_summary_model_denied_when_end_user_over_model_budget():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.model_max_budget_limiter", limiter),
@@ -1743,12 +1743,12 @@ async def test_summary_model_allowed_when_within_model_budget():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.model_max_budget_limiter", limiter),
@@ -1834,12 +1834,12 @@ async def test_summary_model_rate_limit_check_errors(limiter_error, summary_call
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging),
@@ -1876,12 +1876,12 @@ async def test_summary_model_denied_when_over_rate_limit():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging),
@@ -1912,12 +1912,12 @@ async def test_summary_model_allowed_when_within_rate_limit():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging),
@@ -1960,12 +1960,12 @@ async def test_summary_model_allowed_while_the_caller_holds_the_keys_only_parall
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.proxy_logging_obj", _proxy_logging_like_the_live_proxy(limiter)),
@@ -1996,12 +1996,12 @@ async def test_summary_model_rate_limit_skipped_for_legacy_limiter():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch("litellm.proxy.proxy_server.proxy_logging_obj", proxy_logging),
@@ -2050,12 +2050,12 @@ async def test_summary_model_denied_when_team_over_model_budget():
 
     with (
         patch(  # test-quality-ok: apply_compact_20260112 reads the summary model setting as a module global, no seam
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),  # test-quality-ok: forces the over-threshold branch
         patch(  # test-quality-ok: the summary call is the observable that must NOT happen when the team is over budget
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
         patch(  # test-quality-ok: the limiter is a proxy_server module global the editor imports, no injection seam
@@ -2108,12 +2108,12 @@ async def test_scoped_budget_metadata_propagated_to_summary_call():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             return_value=mock_response,
         ) as mock_call,
@@ -2136,7 +2136,7 @@ async def test_scoped_budget_metadata_propagated_to_summary_call():
 async def test_summary_call_passes_end_user_id_as_top_level_user():
     """``_call_summary_model`` forwards the propagated end-user id as the top-level
     ``user`` kwarg that legacy limiter / prometheus end-user tracking reads."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -2159,7 +2159,7 @@ async def test_summary_call_passes_end_user_id_as_top_level_user():
 
 async def test_summary_call_omits_user_when_no_end_user_id():
     """No end-user id on the parent request means no ``user`` kwarg is sent."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -2195,12 +2195,12 @@ async def test_model_budget_metadata_propagated_to_summary_call():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             new_callable=AsyncMock,
             return_value=mock_response,
         ) as mock_call,
@@ -2236,12 +2236,12 @@ async def test_summary_call_propagates_allowed_model_region():
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._call_summary_model",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._call_summary_model",
             mock_call,
         ),
     ):
@@ -2262,7 +2262,7 @@ async def test_summary_call_omits_allowed_model_region_when_unset():
     """Callers without a region restriction must not get an ``allowed_model_region=None``
     kwarg, which would otherwise force the router to evaluate region filtering.
     """
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -2285,7 +2285,7 @@ async def test_summary_call_omits_allowed_model_region_when_unset():
 
 async def test_summary_call_forwards_allowed_model_region_when_set():
     """When the caller is region-restricted, the kwarg reaches the router."""
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _call_summary_model,
     )
 
@@ -2316,7 +2316,7 @@ async def test_dispatcher_routes_compact_edit():
     """compact_20260112 in the dispatcher resolves to opt-in gate when no model set."""
     messages = _simple_messages()
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await apply_context_management(
@@ -2358,7 +2358,7 @@ async def test_dispatcher_trigger_below_minimum_raises_through():
 
 async def test_run_polyfill_skipped_when_context_management_in_additional_drop_params():
     """additional_drop_params=["context_management"] is the explicit opt-out."""
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         _run_polyfill_if_enabled,
     )
 
@@ -2379,13 +2379,13 @@ async def test_run_polyfill_runs_when_litellm_drop_params_true(monkeypatch):
     """drop_params must not disable the polyfill: context_management is a
     LiteLLM-supported param (polyfilled where not native), and drop_params only
     exists to strip genuinely unsupported params."""
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         _run_polyfill_if_enabled,
     )
 
     monkeypatch.setattr(litellm, "drop_params", True)
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+        "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
         return_value=None,
     ):
         result = await _run_polyfill_if_enabled(
@@ -2404,7 +2404,7 @@ async def test_run_polyfill_runs_when_litellm_drop_params_true(monkeypatch):
 
 async def test_run_polyfill_skipped_when_spec_empty():
     """Empty context_management_spec must also return None (no polyfill work)."""
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         _run_polyfill_if_enabled,
     )
 
@@ -2473,7 +2473,7 @@ def _openai_chat_response():
 
 
 async def _call_async_adapter_handler(**handler_kwargs: Any):
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         LiteLLMMessagesToCompletionTransformationHandler,
     )
 
@@ -2532,7 +2532,7 @@ async def test_async_handler_additional_drop_params_strips_context_management():
 
 
 def _call_sync_adapter_handler(**handler_kwargs: Any):
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         LiteLLMMessagesToCompletionTransformationHandler,
     )
 
@@ -2584,7 +2584,7 @@ async def test_prepare_context_managed_request_forwards_proxy_litellm_metadata()
     Anthropic-shape ``metadata`` arg (which only carries ``user_id``). Otherwise
     the summary subcall lands on the router with no parent attribution, and
     those tokens go unbilled to the caller's key/team."""
-    from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+    from litellm.llms.anthropic.pass_through.adapters.handler import (
         _prepare_context_managed_request,
     )
 
@@ -2597,7 +2597,7 @@ async def test_prepare_context_managed_request_forwards_proxy_litellm_metadata()
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact._read_summary_model_setting",
+            "litellm.llms.anthropic.pass_through.context_management.editors.compact._read_summary_model_setting",
             return_value="claude-haiku-4-5",
         ),
         patch("litellm.token_counter", return_value=200_000),
@@ -2786,7 +2786,7 @@ def test_endpoint_runs_failure_hook_on_500_context_management_error():
 
 
 def test_count_effective_tokens_counts_midturn_system_correction():
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _count_effective_tokens,
     )
 
@@ -2813,7 +2813,7 @@ def test_count_effective_tokens_counts_midturn_system_correction():
 
 
 def test_build_summary_messages_keeps_midturn_system_correction_in_place():
-    from litellm.llms.anthropic.experimental_pass_through.context_management.editors.compact import (
+    from litellm.llms.anthropic.pass_through.context_management.editors.compact import (
         _build_summary_messages,
     )
 
@@ -2847,7 +2847,7 @@ async def test_threshold_check_counts_tokens_off_the_event_loop(monkeypatch):
         warm_tokenizer,
     )
 
-    from litellm.llms.anthropic.experimental_pass_through.context_management.constants import (
+    from litellm.llms.anthropic.pass_through.context_management.constants import (
         COMPACT_SUMMARY_MODEL_SETTING_KEY,
     )
     from litellm.proxy.proxy_server import general_settings

tests/unit/llms/anthropic/pass_through/context_management/test_dispatcher.py

@@ -2,7 +2,7 @@
 Unit tests for the context_management polyfill dispatcher.
 """
 
-from litellm.llms.anthropic.experimental_pass_through.context_management import (
+from litellm.llms.anthropic.pass_through.context_management import (
     apply_context_management,
 )
 

tests/unit/llms/anthropic/pass_through/messages/test_advisor_integration.py

@@ -72,7 +72,7 @@ async def test_full_dispatch_interceptor_fires_and_loop_completes():
     The interceptor must fire, run the loop (1 advisor call), and return a
     clean final response with no advisor tool_use blocks.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages,
     )
 
@@ -88,7 +88,7 @@ async def test_full_dispatch_interceptor_fires_and_loop_completes():
         return _text_resp("def is_prime(n): ...")  # executor: final
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_handler,
     ):
         result = await anthropic_messages(
@@ -127,10 +127,10 @@ async def test_max_uses_enforced_through_full_handler():
     AdvisorMaxIterationsError propagates out of anthropic_messages() when
     the executor keeps calling the advisor past max_uses.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages,
     )
-    from litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor import (
+    from litellm.llms.anthropic.pass_through.messages.interceptors.advisor import (
         AdvisorMaxIterationsError,
     )
 
@@ -143,7 +143,7 @@ async def test_max_uses_enforced_through_full_handler():
         return _advisor_call_resp()
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_handler,
     ):
         with pytest.raises(AdvisorMaxIterationsError):
@@ -168,7 +168,7 @@ async def test_anthropic_provider_bypasses_interceptor():
     With custom_llm_provider='anthropic', the interceptor must NOT fire.
     The advisor_20260301 tool is forwarded as-is to the underlying handler.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages,
     )
 
@@ -176,7 +176,7 @@ async def test_anthropic_provider_bypasses_interceptor():
 
     # Patch the non-interceptor code path — anthropic_messages_handler
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.handler.anthropic_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.handler.anthropic_messages_handler",
         return_value=direct_response,
     ) as mock_native:
         result = await anthropic_messages(
@@ -215,7 +215,7 @@ async def test_named_params_forwarded_into_advisor_executor_subcall():
     them, e.g. Vertex AI rejecting ``clear_thinking_20251015`` context_management
     edits with: ``strategy requires thinking to be enabled or adaptive``.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages,
     )
 
@@ -246,7 +246,7 @@ async def test_named_params_forwarded_into_advisor_executor_subcall():
         return _text_resp("Final answer.")
 
     with patch(
-        "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+        "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
         side_effect=mock_handler,
     ):
         await anthropic_messages(
@@ -298,7 +298,7 @@ async def test_pre_request_hook_override_does_not_collide_with_explicit_kwargs()
 
     Regression for Greptile P2 on PR #27810.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages,
     )
 
@@ -339,11 +339,11 @@ async def test_pre_request_hook_override_does_not_collide_with_explicit_kwargs()
 
     with (
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.handler._execute_pre_request_hooks",
+            "litellm.llms.anthropic.pass_through.messages.handler._execute_pre_request_hooks",
             side_effect=fake_pre_request_hooks,
         ),
         patch(
-            "litellm.llms.anthropic.experimental_pass_through.messages.interceptors.advisor._call_messages_handler",
+            "litellm.llms.anthropic.pass_through.messages.interceptors.advisor._call_messages_handler",
             side_effect=mock_handler,
         ),
     ):

tests/unit/llms/anthropic/pass_through/messages/test_agentic_streaming_iterator.py

@@ -11,7 +11,7 @@ import pytest
 
 
 from litellm.constants import STREAM_SSE_KEEPALIVE_PING_BYTES
-from litellm.llms.anthropic.experimental_pass_through.messages.agentic_streaming_iterator import (
+from litellm.llms.anthropic.pass_through.messages.agentic_streaming_iterator import (
     SERVER_FULFILLED_TOOL_LEAK_ERROR_SSE_BYTES,
     AgenticAnthropicStreamingIterator,
     _handle_content_block_delta,

tests/unit/llms/anthropic/pass_through/messages/test_anthropic_experimental_pass_through_messages_handler.py

@@ -30,7 +30,7 @@ def test_anthropic_experimental_pass_through_messages_handler():
     Test that api key is passed to litellm.responses for OpenAI models.
     OpenAI and Azure models are routed directly to the Responses API.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -115,7 +115,7 @@ def test_anthropic_experimental_pass_through_messages_handler_dynamic_api_key_an
     Test that api key, api base, and extra kwargs are forwarded to litellm.completion for Azure models.
     Azure models are routed through chat/completions (not the Responses API).
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -142,7 +142,7 @@ async def test_anthropic_messages_sanitizes_empty_text_blocks_before_dispatch():
     """Regression test for #22930.  The unified /v1/messages path must
     strip empty text blocks before forwarding, otherwise Anthropic
     returns 400 "text content blocks must be non-empty"."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     msgs = [
         {
@@ -180,7 +180,7 @@ async def test_anthropic_messages_sanitizes_empty_text_blocks_before_dispatch():
 
 @pytest.mark.asyncio
 async def test_anthropic_messages_sanitizes_tool_use_ids_before_dispatch():
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     msgs = [
         {
@@ -231,7 +231,7 @@ def test_anthropic_experimental_pass_through_messages_handler_custom_llm_provide
     Provider resolution now happens exactly once, inside litellm.completion itself
     (BerriAI/litellm#37716), so the handler passes the original unresolved model through.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -315,7 +315,7 @@ def test_openai_model_with_thinking_converts_to_reasoning():
     OpenAI models are routed directly to the Responses API, so we verify that
     litellm.responses() is called with `reasoning` properly set.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -355,7 +355,7 @@ class TestThinkingParameterTransformation:
 
     def test_claude_model_preserves_thinking_with_budget_tokens(self):
         """Test that Claude models get thinking parameter passed through with exact budget_tokens."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -370,7 +370,7 @@ class TestThinkingParameterTransformation:
 
     def test_non_claude_model_converts_thinking_to_reasoning_effort(self):
         """Test that non-Claude models convert thinking to reasoning_effort."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -388,7 +388,7 @@ class TestThinkingParameterTransformation:
     def test_translate_thinking_for_model_summary_when_enabled(self):
         """When reasoning_auto_summary is True, summary='detailed' is injected."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -406,7 +406,7 @@ class TestThinkingParameterTransformation:
 
     def test_translate_thinking_for_model_preserves_user_summary(self):
         """User-provided summary is always preserved regardless of flag."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -423,7 +423,7 @@ class TestThinkingSummaryPreservation:
 
     def test_thinking_summary_concise_preserved_for_openai(self):
         """User-provided summary='concise' should not be replaced with 'detailed'."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -439,7 +439,7 @@ class TestThinkingSummaryPreservation:
 
     def test_thinking_summary_auto_preserved_for_openai(self):
         """User-provided summary='auto' should be preserved."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -456,7 +456,7 @@ class TestThinkingSummaryPreservation:
     def test_summary_added_when_auto_summary_enabled(self):
         """When reasoning_auto_summary is True, summary='detailed' is added."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -481,7 +481,7 @@ class TestThinkingSummaryPreservation:
     def test_no_summary_by_default_string_reasoning(self):
         """By default (reasoning_auto_summary=False), summary is not added for string reasoning_effort."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -504,7 +504,7 @@ class TestThinkingSummaryPreservation:
     def test_no_summary_by_default_dict_reasoning(self):
         """By default (reasoning_auto_summary=False), summary is not injected into dict reasoning_effort."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -527,7 +527,7 @@ class TestThinkingSummaryPreservation:
     def test_summary_added_when_env_var_set(self, monkeypatch):
         """When LITELLM_REASONING_AUTO_SUMMARY env var is true, summary is added."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -554,7 +554,7 @@ class TestThinkingSummaryPreservation:
     def test_user_provided_summary_preserved_even_when_flag_off(self):
         """When user already set summary in dict reasoning_effort, it's preserved regardless of flag."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
@@ -575,7 +575,7 @@ class TestThinkingSummaryPreservation:
 
     def test_openai_model_with_thinking_summary_end_to_end(self):
         """End-to-end: anthropic_messages_handler should preserve thinking.summary for OpenAI models."""
-        from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+        from litellm.llms.anthropic.pass_through.messages.handler import (
             anthropic_messages_handler,
         )
 
@@ -604,7 +604,7 @@ class TestThinkingSummaryPreservation:
 
     def test_responses_adapter_preserves_summary(self):
         """translate_thinking_to_reasoning should include summary when user provides it."""
-        from litellm.llms.anthropic.experimental_pass_through.responses_adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.responses_adapters.transformation import (
             LiteLLMAnthropicToResponsesAPIAdapter,
         )
 
@@ -615,7 +615,7 @@ class TestThinkingSummaryPreservation:
     def test_responses_adapter_no_summary_by_default(self):
         """translate_thinking_to_reasoning should not include summary by default (opt-in)."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.responses_adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.responses_adapters.transformation import (
             LiteLLMAnthropicToResponsesAPIAdapter,
         )
 
@@ -631,7 +631,7 @@ class TestThinkingSummaryPreservation:
 
     def test_translate_thinking_for_model_preserves_summary(self):
         """translate_thinking_for_model should include summary in reasoning_effort dict when user provides it."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -645,7 +645,7 @@ class TestThinkingSummaryPreservation:
     def test_translate_thinking_for_model_disabled_stays_plain_string_when_auto_summary_enabled(self):
         """Disabled thinking must stay a plain string even when reasoning_auto_summary is on."""
         import litellm
-        from litellm.llms.anthropic.experimental_pass_through.adapters.transformation import (
+        from litellm.llms.anthropic.pass_through.adapters.transformation import (
             LiteLLMAnthropicMessagesAdapter,
         )
 
@@ -684,7 +684,7 @@ def _empty_block_msgs():
 
 def test_handler_strips_when_no_presanitized_flag():
     """Sync entry point (no async wrapper): handler must still sanitize."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     with patch.object(
         handler,
@@ -704,7 +704,7 @@ def test_handler_strips_when_no_presanitized_flag():
 
 def test_handler_skips_strip_when_presanitized():
     """Async wrapper already sanitized -> handler must NOT rescan."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     with patch.object(
         handler,
@@ -725,7 +725,7 @@ def test_handler_skips_strip_when_presanitized():
 
 def test_handler_flattens_replayed_unencrypted_web_search_results():
     """Synthesized search blocks replayed as history must reach the provider as text."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     captured = {}
 
@@ -780,7 +780,7 @@ def test_handler_flattens_replayed_unencrypted_web_search_results():
 
 def test_presanitized_flag_not_leaked_to_provider_params():
     """The private sentinel must be popped, never forwarded as a request param."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     captured = {}
 
@@ -809,7 +809,7 @@ def test_presanitized_flag_not_leaked_to_provider_params():
 @pytest.mark.asyncio
 async def test_async_wrapper_sets_presanitized_and_sanitizes_once():
     """End-to-end: wrapper sanitizes (once) AND signals the handler to skip."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     captured = {}
 
@@ -853,7 +853,7 @@ def _gate_stubs(monkeypatch):
     provider config handed to the native passthrough path and ``translation_calls``
     counts hits on the Anthropic->OpenAI translation handlers.
     """
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     captured = {}
     translation_calls = {"count": 0}
@@ -883,7 +883,7 @@ def _gate_stubs(monkeypatch):
 def test_gate_passthrough_when_supported_endpoints_opts_in(monkeypatch):
     """provider=openai + model_info.supported_endpoints containing /v1/messages
     must route to the native passthrough config, NOT the translation handlers."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
     from litellm.llms.openai_like.messages.transformation import (
@@ -909,7 +909,7 @@ def test_gate_passthrough_when_supported_endpoints_opts_in(monkeypatch):
 def test_gate_translates_when_supported_endpoints_absent(monkeypatch):
     """Default behavior is unchanged: without the /v1/messages opt-in, an openai
     deployment is translated (Responses API), never passed through natively."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -931,7 +931,7 @@ def test_gate_translates_when_supported_endpoints_absent(monkeypatch):
 def test_gate_passthrough_skipped_when_only_chat_completions_supported(monkeypatch):
     """A deployment that lists only /v1/chat/completions is still translated;
     the opt-in is specifically the /v1/messages entry."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -964,7 +964,7 @@ def test_gate_passthrough_forwards_cache_control_ttl_only_when_deployment_opts_i
 ):
     """The passthrough config strips cache_control.ttl unless the deployment sets
     model_info.cache_control_ttl to exactly true."""
-    from litellm.llms.anthropic.experimental_pass_through.messages.handler import (
+    from litellm.llms.anthropic.pass_through.messages.handler import (
         anthropic_messages_handler,
     )
 
@@ -1275,7 +1275,7 @@ class TestMessagesStreamingSuccessLogging:
     @pytest.mark.asyncio
     async def test_responses_bridge_streaming_emits_success_logging(self, capture_success_payloads):
         """The Responses bridge, which is the default for openai/ deployments."""
-        from litellm.llms.anthropic.experimental_pass_through.responses_adapters.handler import (
+        from litellm.llms.anthropic.pass_through.responses_adapters.handler import (
             LiteLLMMessagesToResponsesAPIHandler,
         )
 
@@ -1316,14 +1316,14 @@ class TestMessagesStreamingSuccessLogging:
         """The chat-completions bridge, reached via
         litellm.use_chat_completions_url_for_anthropic_messages. Its router lookup is
         stubbed to what an SDK caller with no proxy running already resolves to."""
-        from litellm.llms.anthropic.experimental_pass_through.adapters.handler import (
+        from litellm.llms.anthropic.pass_through.adapters.handler import (
             LiteLLMMessagesToCompletionTransformationHandler,
         )
 
         _bind_logging_worker_to_running_loop()
 
         with patch(
-            "litellm.llms.anthropic.experimental_pass_through.adapters.handler._proxy_router_fallback",
+            "litellm.llms.anthropic.pass_through.adapters.handler._proxy_router_fallback",
             return_value=None,
         ):
             sse_stream = await LiteLLMMessagesToCompletionTransformationHandler.async_anthropic_messages_handler(
@@ -1376,7 +1376,7 @@ async def test_anthropic_messages_maps_provider_exception_before_failure_logging
     The 403 row pins the upstream status on the way through the mapper: Anthropic's
     documented permission_error must reach the caller as a 403, never as the mapper's
     APIConnectionError 500 fallthrough."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     capture = _FailureCapture()
     monkeypatch.setattr(litellm, "callbacks", [capture])
@@ -1418,7 +1418,7 @@ async def test_anthropic_messages_leaves_non_provider_failures_unmapped():
     """The mapping boundary is for provider failures only. A request rejected before
     the provider call (here invalid metadata) must surface as the original exception,
     not as the mapper's APIConnectionError, whose message embeds a server traceback."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     def upstream_must_not_be_called(request: httpx.Request) -> httpx.Response:
         raise AssertionError("the provider must not be called for a request rejected locally")
@@ -1464,7 +1464,7 @@ def _recording_client(seen_urls: list[str]) -> AsyncHTTPHandler:
 
 @pytest.mark.asyncio
 async def test_provider_messages_api_base_env_is_not_shadowed_by_the_chat_default(monkeypatch):
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     monkeypatch.delenv("DEEPSEEK_API_BASE", raising=False)
     monkeypatch.setenv("DEEPSEEK_ANTHROPIC_API_BASE", "https://deepseek.internal.example/anthropic")
@@ -1483,7 +1483,7 @@ async def test_provider_messages_api_base_env_is_not_shadowed_by_the_chat_defaul
 @pytest.mark.asyncio
 async def test_anthropic_messages_forwards_safeguards_and_unknown_beta_to_anthropic():
     """Shapes are what Claude Code 2.1.278 sends and api.anthropic.com returns, captured 2026-09-21."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     safeguards = [{"type": "dangerous_tool_use", "classifier_context": {"v": 1, "permission_mode": "auto"}}]
     client_betas = "dangerous-tool-use-2026-09-03,interleaved-thinking-2025-05-14"
@@ -1531,7 +1531,7 @@ async def test_anthropic_messages_forwards_safeguards_and_unknown_beta_to_anthro
 @pytest.mark.asyncio
 async def test_anthropic_messages_streaming_forwards_safeguards_and_keeps_safeguard_results():
     """Shapes are what Claude Code 2.1.278 sends and api.anthropic.com returns, captured 2026-09-21."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     safeguards = [{"type": "dangerous_tool_use", "classifier_context": {"v": 1, "permission_mode": "auto"}}]
     tool_verdicts = {"toolu_01": {"type": "evaluated", "outcome": "not_flagged"}}
@@ -1632,7 +1632,7 @@ async def test_anthropic_messages_forwards_safeguards_and_dangerous_tool_use_bet
     local_beta_headers_config, client_headers
 ):
     """Bedrock Invoke takes betas in the body's `anthropic_beta` and 400s on `safeguards` without the beta, so the beta rides along with the field."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
 
     safeguards, safeguard_results = _claude_code_auto_mode_request()
     captured: dict[str, object] = {}
@@ -1661,7 +1661,7 @@ async def test_anthropic_messages_forwards_safeguards_and_dangerous_tool_use_bet
     local_beta_headers_config, client_headers
 ):
     """Vertex rawPredict takes the beta as the `anthropic-beta` header and 400s on `safeguards` without it, so the beta rides along with the field."""
-    from litellm.llms.anthropic.experimental_pass_through.messages import handler
+    from litellm.llms.anthropic.pass_through.messages import handler
     from litellm.llms.vertex_ai.vertex_llm_base import VertexBase
 
     safeguards, safeguard_results = _claude_code_auto_mode_request()

tests/unit/llms/anthropic/pass_through/messages/test_anthropic_messages_effort.py

@@ -7,7 +7,7 @@ from litellm.constants import (
     DEFAULT_REASONING_EFFORT_XHIGH_THINKING_BUDGET,
 )
 from litellm.llms.anthropic.common_utils import AnthropicError
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.openai_like.json_loader import SimpleProviderConfig

tests/unit/llms/anthropic/pass_through/messages/test_anthropic_messages_encrypted_reasoning.py

@@ -1,7 +1,7 @@
 from litellm.litellm_core_utils.prompt_templates.common_utils import (
     encrypted_reasoning_signature,
 )
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 

tests/unit/llms/anthropic/pass_through/messages/test_anthropic_messages_per_turn_control.py

@@ -2,7 +2,7 @@ import pytest
 
 from litellm import anthropic_beta_headers_manager
 from litellm.anthropic_beta_headers_manager import update_headers_with_filtered_beta
-from litellm.llms.anthropic.experimental_pass_through.messages.transformation import (
+from litellm.llms.anthropic.pass_through.messages.transformation import (
     AnthropicMessagesConfig,
 )
 from litellm.llms.openai_like.json_loader import SimpleProviderConfig