fix(bedrock): route model_id overrides to Converse and never send an empty bearer natively

A deployment whose litellm_params carry model_id (an application inference profile or provisioned throughput ARN) went to the native Chat Completions route with the base model in the URL and model_id left in the body. It now takes Converse like the bedrock/arn:... model form, which encodes the override into the request URL

A blank api_key on a SigV4 deployment became an Authorization header reading Bearer with nothing after it on the native route, since the OpenAI-like header builder writes any non-None key and the signer keeps a non-AWS4 Authorization header. validate_environment now resolves the key through bedrock_bearer_token, so a blank key is signed with SigV4 the way Converse signs it

litellm/llms/bedrock/chat/chat_completions/transformation.py

@@ -31,7 +31,7 @@ from litellm.litellm_core_utils.prompt_templates.image_handling import (
     inline_remote_media,
 )
 from litellm.llms.base_llm.chat.transformation import BaseLLMException
-from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM
+from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM, bedrock_bearer_token
 from litellm.llms.bedrock.common_utils import (
     BedrockError,
     bedrock_model_is_openai_gpt,
@@ -263,6 +263,26 @@ class AmazonBedrockRuntimeChatCompletionsConfig(OpenAILikeChatConfig):
     ) -> BaseLLMException:
         return BedrockError(status_code=status_code, message=error_message, headers=headers)
 
+    def validate_environment(
+        self,
+        headers: dict,  # mutable-ok: BaseConfig signature
+        model: str,
+        messages: list[AllMessageValues],
+        optional_params: dict,  # mutable-ok: BaseConfig signature
+        litellm_params: dict,  # mutable-ok: BaseConfig signature
+        api_key: str | None = None,
+        api_base: str | None = None,
+    ) -> dict:  # mutable-ok: BaseConfig signature
+        return super().validate_environment(
+            headers=headers,
+            model=model,
+            messages=messages,
+            optional_params=optional_params,
+            litellm_params=litellm_params,
+            api_key=bedrock_bearer_token(api_key),
+            api_base=api_base,
+        )
+
     def get_complete_url(
         self,
         api_base: str | None,

litellm/llms/bedrock/common_utils.py

@@ -906,6 +906,7 @@ BEDROCK_CONVERSE_ONLY_REQUEST_KEYS: Final = frozenset(
         "additionalModelRequestFields",
         "top_k",
         "stop",
+        "model_id",
     )
 )
 
@@ -931,7 +932,9 @@ def bedrock_request_needs_converse(model: str, request_params: Mapping[str, obje
     Converse-shaped body keys (``BEDROCK_CONVERSE_ONLY_REQUEST_KEYS``, the Anthropic-style ``thinking``
     block and the ``additionalModelRequestFields`` / ``top_k`` extension params included, which only Converse
     forwards as ``additionalModelRequestFields`` and ``inferenceConfig``) have no field on
-    AWS's native OpenAI surface, ``stop`` stays on Converse where it fails loudly instead of silently
+    AWS's native OpenAI surface, a ``model_id`` override (an application inference profile or provisioned
+    throughput ARN) is only encoded into Converse's request URL and so stays on Converse like the
+    ``bedrock/arn:...`` model form, ``stop`` stays on Converse where it fails loudly instead of silently
     stopping hidden reasoning, operator-owned request metadata is only written onto the Converse body,
     function tools (``tools`` or legacy ``functions``) on a model without
     ``supports_bedrock_runtime_chat_completions_tools_with_reasoning`` are rejected there unless

tests/unit/llms/bedrock/chat/chat_completions/test_bedrock_chat_completions_transformation.py

@@ -25,6 +25,8 @@ from litellm.llms.bedrock.common_utils import (
 )
 from litellm.llms.custom_httpx.http_handler import HTTPHandler
 
+APPLICATION_INFERENCE_PROFILE_ARN = "arn:aws:bedrock:us-west-2:123412341234:application-inference-profile/a1b2c3"
+
 
 @pytest.fixture
 def local_cost_map(monkeypatch):
@@ -425,8 +427,13 @@ def test_guardrail_config_falls_back_to_converse(local_cost_map, model):
 )
 @pytest.mark.parametrize(
     "request_params",
-    [{"additionalModelRequestFields": {"reasoning_effort": "high"}}, {"top_k": 40}, {"stop": ["END"]}],
-    ids=["additionalModelRequestFields", "top_k", "stop"],
+    [
+        {"additionalModelRequestFields": {"reasoning_effort": "high"}},
+        {"top_k": 40},
+        {"stop": ["END"]},
+        {"model_id": APPLICATION_INFERENCE_PROFILE_ARN},
+    ],
+    ids=["additionalModelRequestFields", "top_k", "stop", "model_id"],
 )
 def test_converse_extension_params_fall_back_to_converse(local_cost_map, model, request_params):
     assert bedrock_request_needs_converse(model, request_params) is True
@@ -434,6 +441,59 @@ def test_converse_extension_params_fall_back_to_converse(local_cost_map, model,
     assert BedrockModelInfo.get_bedrock_route(model, {key: None for key in request_params}) == "chat_completions"
 
 
+@pytest.mark.parametrize(
+    "model", ["bedrock/us.openai.gpt-5.6-sol", "global.openai.gpt-6-sol", "bedrock/chat_completions/us.xai.grok-4.6"]
+)
+def test_model_id_override_is_served_by_converse_like_the_arn_model_form(local_cost_map, model):
+    assert bedrock_route_for_request(model, {"model_id": APPLICATION_INFERENCE_PROFILE_ARN}, None) == "converse"
+    assert bedrock_route_for_request(model, {"model_id": None}, None) == "chat_completions"
+
+
+SIGV4_PARAMS = {
+    "aws_access_key_id": "AKIAIOSFODNN7EXAMPLE",
+    "aws_secret_access_key": "wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY",
+    "aws_region_name": "us-east-1",
+}
+
+
+@pytest.mark.parametrize("api_key", ["", None], ids=["blank", "absent"])
+def test_blank_api_key_is_signed_with_sigv4_instead_of_an_empty_bearer(monkeypatch, api_key):
+    monkeypatch.delenv("AWS_BEARER_TOKEN_BEDROCK", raising=False)
+    cfg = AmazonBedrockRuntimeChatCompletionsConfig()
+    url = "https://bedrock-runtime.us-east-1.amazonaws.com/openai/v1/chat/completions"
+    headers = cfg.validate_environment(
+        headers={},
+        model="bedrock/us.openai.gpt-5.6-sol",
+        messages=[{"role": "user", "content": "hello"}],
+        optional_params=dict(SIGV4_PARAMS),
+        litellm_params={},
+        api_key=api_key,
+    )
+    assert "Authorization" not in headers
+    signed, _ = cfg.sign_request(
+        headers=headers,
+        optional_params=dict(SIGV4_PARAMS),
+        request_data={"model": "us.openai.gpt-5.6-sol", "messages": []},
+        api_base=url,
+        api_key=api_key,
+    )
+    assert signed["Authorization"].startswith("AWS4-HMAC-SHA256 Credential=AKIAIOSFODNN7EXAMPLE/"), signed
+
+
+def test_bearer_api_key_is_sent_as_the_authorization_header(monkeypatch):
+    monkeypatch.delenv("AWS_BEARER_TOKEN_BEDROCK", raising=False)
+    cfg = AmazonBedrockRuntimeChatCompletionsConfig()
+    headers = cfg.validate_environment(
+        headers={},
+        model="bedrock/us.openai.gpt-5.6-sol",
+        messages=[{"role": "user", "content": "hello"}],
+        optional_params={},
+        litellm_params={},
+        api_key="bedrock-api-key",
+    )
+    assert headers["Authorization"] == "Bearer bedrock-api-key"
+
+
 @pytest.mark.parametrize(
     "request_params, expected_route",
     [