fix(bedrock): answer 400, not 500, for malformed count-tokens requests (#50708)

* fix(bedrock): answer 400, not 500, for malformed native count-tokens requests

The Bedrock count-tokens handler now checks the body before any AWS call and raises a 400 BedrockError for a missing or empty messages list, a message without role or content, or a missing model. The token counter marks that rejection as local, so it skips the Mantle fallback and the framing estimate, and /v1/messages/count_tokens answers invalid_request_error

get_bedrock_base_model returns a bare cross-region prefix such as us or global unchanged instead of raising IndexError, so Bedrock judges the model id and its own 400 reaches the caller

* fix(bedrock): answer 400 for a malformed message on Mantle token counts

* fix(bedrock): limit the region-only model id fix to token counting

* fix(bedrock): count a routing-prefixed region-only id and leave other providers' count errors as they were

---------

Co-authored-by: mateo-berri <277851410+mateo-berri@users.noreply.github.com>

litellm/llms/bedrock/count_tokens/bedrock_token_counter.py

@@ -21,7 +21,7 @@ from litellm.llms.bedrock.common_utils import (
     BedrockError,
     get_bedrock_base_model,
 )
-from litellm.llms.bedrock.count_tokens.handler import BedrockCountTokensHandler
+from litellm.llms.bedrock.count_tokens.handler import BedrockCountTokensHandler, BedrockCountTokensRequestError
 from litellm.llms.bedrock.count_tokens.mantle_handler import BedrockMantleCountTokensHandler
 from litellm.llms.bedrock_mantle.common_utils import is_mantle_claude_model
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
@@ -54,6 +54,7 @@ class CountTokensFailure:
     status_code: int
     message: str
     tokenizer_type: str
+    rejected_before_provider_call: bool = False
 
 
 CountTokensOutcome = CountedTokens | CountTokensFailure
@@ -74,6 +75,7 @@ def _claude_count_translator(litellm_params: Mapping[str, object]) -> AnthropicC
 def _runtime_rejected_claude_model(outcome: CountTokensOutcome, resolved_model: str) -> bool:
     return (
         isinstance(outcome, CountTokensFailure)
+        and not outcome.rejected_before_provider_call
         and outcome.status_code in _RUNTIME_FALLBACK_STATUSES
         and is_mantle_claude_model(resolved_model)
     )
@@ -162,6 +164,13 @@ class BedrockTokenCounter(BaseTokenCounter):
                 client=self._client,
             )
             reply: Final = _CountTokensReply.model_validate(result)
+        except BedrockCountTokensRequestError as e:
+            return CountTokensFailure(
+                status_code=e.status_code,
+                message=e.message,
+                tokenizer_type=tokenizer_type,
+                rejected_before_provider_call=True,
+            )
         except BedrockError as e:
             verbose_logger.debug(
                 "%s CountTokens API error: status=%s, message=%s", tokenizer_type, e.status_code, e.message
@@ -268,6 +277,7 @@ class BedrockTokenCounter(BaseTokenCounter):
                     error=True,
                     error_message=outcome.message,
                     status_code=outcome.status_code,
+                    rejected_before_provider_call=outcome.rejected_before_provider_call,
                 )
             case _:
                 assert_never(outcome)

litellm/llms/bedrock/count_tokens/handler.py

@@ -20,6 +20,11 @@ from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, get_async_h
 _DECODED_JSON: Final = TypeAdapter(object)
 
 
+class BedrockCountTokensRequestError(BedrockError):
+    def __init__(self, problem: str) -> None:
+        super().__init__(status_code=400, message=problem)
+
+
 class BedrockCountTokensHandler(BedrockCountTokensConfig):
     """
     Simplified handler for AWS Bedrock CountTokens API requests.
@@ -45,10 +50,10 @@ class BedrockCountTokensHandler(BedrockCountTokensConfig):
         Returns:
             Dictionary containing token count response
         """
+        problem: Final = self.count_tokens_request_problem(request_data)
+        if problem is not None:
+            raise BedrockCountTokensRequestError(problem)
         try:
-            # Validate the request
-            self.validate_count_tokens_request(request_data)
-
             verbose_logger.debug("Processing CountTokens request for resolved model: %s", resolved_model)
 
             # Get AWS region using existing LiteLLM function

litellm/llms/bedrock/count_tokens/mantle_handler.py

@@ -12,6 +12,8 @@ from litellm.llms.anthropic.common_utils import context_management_betas, tools_
 from litellm.llms.anthropic.count_tokens.transformation import COUNT_TOKEN_OPTION_NAMES, AnthropicCountTokensConfig
 from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM, run_aws_signing
 from litellm.llms.bedrock.common_utils import BedrockError, build_mantle_messages_url
+from litellm.llms.bedrock.count_tokens.handler import BedrockCountTokensRequestError
+from litellm.llms.bedrock.count_tokens.transformation import count_tokens_messages_problem
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler, get_async_httpx_client
 
 MANTLE_COUNT_TOKENS_SUFFIX: Final = "/count_tokens"
@@ -58,6 +60,9 @@ class BedrockMantleCountTokensHandler(BaseAWSLLM):
     ) -> dict[str, JsonValue]:
         try:
             request: Final = _COUNT_REQUEST.validate_python(request_data)
+            problem: Final = count_tokens_messages_problem(request["messages"])
+            if problem is not None:
+                raise BedrockCountTokensRequestError(problem)
             aws_region_name: Final = await run_aws_signing(
                 self.get_aws_region_name, optional_params=litellm_params, model=resolved_model, model_id=None
             )

litellm/llms/bedrock/count_tokens/transformation.py

@@ -6,7 +6,7 @@ to AWS Bedrock's CountTokens API format and vice versa.
 """
 
 import re
-from collections.abc import Mapping
+from collections.abc import Mapping, Sequence
 from typing import Final, Literal
 
 from pydantic import JsonValue
@@ -14,7 +14,11 @@ from pydantic import JsonValue
 from litellm.litellm_core_utils.url_utils import join_api_base
 from litellm.llms.anthropic.common_utils import tools_without_custom_field
 from litellm.llms.bedrock.base_aws_llm import BaseAWSLLM
-from litellm.llms.bedrock.common_utils import get_bedrock_base_model
+from litellm.llms.bedrock.common_utils import (
+    get_bedrock_base_model,
+    get_bedrock_cross_region_inference_regions,
+    strip_bedrock_routing_prefix,
+)
 
 # Placeholder satisfying the Anthropic InvokeModel schema's required
 # max_tokens field; CountTokens only counts input, so it has no effect
@@ -46,6 +50,23 @@ def _to_converse_message(message: JsonValue) -> dict[str, JsonValue]:
     }
 
 
+def _converse_message_problem(index: int, message: JsonValue) -> str | None:
+    if not isinstance(message, dict):
+        return f"Message {index} must be a dictionary"
+    if "role" not in message:
+        return f"Message {index} must have a 'role' field"
+    if "content" not in message and "tool_calls" in message:
+        return f"Message {index} has tool_calls and no content, which token counting cannot convert for this model"
+    if "content" not in message:
+        return f"Message {index} must have a 'content' field"
+    return None
+
+
+def count_tokens_messages_problem(messages: Sequence[JsonValue]) -> str | None:
+    problems: Final = (_converse_message_problem(index, message) for index, message in enumerate(messages))
+    return next((problem for problem in problems if problem is not None), None)
+
+
 def _sanitized_bedrock_tool_name(raw_name: JsonValue) -> str:
     name: Final = re.sub(r"[^a-zA-Z0-9_]", "_", raw_name if isinstance(raw_name, str) else "")
     prefixed: Final = name if not name or name[0].isalpha() else f"t_{name}"
@@ -217,6 +238,9 @@ class BedrockCountTokensConfig(BaseAWSLLM):
 
     @staticmethod
     def count_tokens_model_id(model: str) -> str:
+        model_id: Final = strip_bedrock_routing_prefix(model)
+        if model_id in get_bedrock_cross_region_inference_regions():
+            return model_id
         return get_bedrock_base_model(model).removeprefix("bedrock/")
 
     def get_bedrock_count_tokens_endpoint(
@@ -265,42 +289,12 @@ class BedrockCountTokensConfig(BaseAWSLLM):
 
         return {"input_tokens": input_tokens}
 
-    def validate_count_tokens_request(self, request_data: Mapping[str, JsonValue]) -> None:
-        """
-        Validate the incoming count tokens request.
-        Supports both Converse and InvokeModel input formats.
-
-        Args:
-            request_data: The request payload
-
-        Raises:
-            ValueError: If the request is invalid
-        """
+    def count_tokens_request_problem(self, request_data: Mapping[str, JsonValue]) -> str | None:
         if not request_data.get("model"):
-            raise ValueError("model parameter is required")
-
-        input_type: Final = self._detect_input_type(request_data)
-
-        if input_type == "converse":
-            # Validate Converse format (messages-based)
-            messages: Final = request_data.get("messages", [])
-            if not messages:
-                raise ValueError("messages parameter is required for Converse input")
-
-            if not isinstance(messages, list):
-                raise ValueError("messages must be a list")
-
-            for i, message in enumerate(messages):
-                if not isinstance(message, dict):
-                    raise ValueError(f"Message {i} must be a dictionary")
-
-                if "role" not in message:
-                    raise ValueError(f"Message {i} must have a 'role' field")
-
-                if "content" not in message:
-                    raise ValueError(f"Message {i} must have a 'content' field")
-        else:
-            # For InvokeModel format, we need at least some content to count tokens
-            # The content structure varies by model, so we do minimal validation
-            if len(request_data) <= 1:  # Only has 'model' field
-                raise ValueError("Request must contain content to count tokens")
+            return "model parameter is required"
+        if self._detect_input_type(request_data) == "invokeModel":
+            return "Request must contain content to count tokens" if len(request_data) <= 1 else None
+        messages: Final = _json_list(request_data.get("messages"))
+        if not messages:
+            return "messages parameter is required for Converse input"
+        return count_tokens_messages_problem(messages)

tests/unit/llms/bedrock/count_tokens/test_bedrock_count_tokens_handler.py

@@ -6,6 +6,7 @@ import httpx
 import pytest
 from botocore.credentials import RefreshableCredentials
 
+from litellm.llms.bedrock.common_utils import BedrockError
 from litellm.llms.bedrock.count_tokens.handler import BedrockCountTokensHandler
 from litellm.llms.bedrock.count_tokens.transformation import BedrockCountTokensConfig
 from litellm.llms.custom_httpx.http_handler import AsyncHTTPHandler
@@ -85,6 +86,42 @@ async def test_handle_count_tokens_request_looks_up_its_region_off_the_event_loo
     assert result == {"input_tokens": 7}
 
 
+_COUNTED_MODEL: Final = "us.anthropic.claude-opus-4-8"
+
+
+@pytest.mark.parametrize(
+    ("request_data", "problem"),
+    [
+        ({"model": _COUNTED_MODEL}, "Request must contain content to count tokens"),
+        ({"model": _COUNTED_MODEL, "messages": []}, "messages parameter is required for Converse input"),
+        ({"model": _COUNTED_MODEL, "messages": [{"role": "user"}]}, "Message 0 must have a 'content' field"),
+        ({"model": _COUNTED_MODEL, "messages": [{"content": "hi"}]}, "Message 0 must have a 'role' field"),
+        ({"model": _COUNTED_MODEL, "messages": ["hi"]}, "Message 0 must be a dictionary"),
+        (
+            {"model": _COUNTED_MODEL, "messages": [{"role": "user", "content": "hi"}, {"role": "assistant"}]},
+            "Message 1 must have a 'content' field",
+        ),
+        ({"messages": [{"role": "user", "content": "hi"}]}, "model parameter is required"),
+    ],
+)
+@pytest.mark.asyncio
+async def test_handle_count_tokens_request_answers_400_for_a_malformed_body_without_calling_bedrock(
+    request_data: dict[str, object], problem: str
+) -> None:
+    client: Final = AsyncMock(spec=AsyncHTTPHandler)
+
+    with pytest.raises(BedrockError) as exc_info:
+        await BedrockCountTokensHandler().handle_count_tokens_request(
+            request_data=request_data,
+            litellm_params={"aws_region_name": "us-east-1"},
+            resolved_model=_COUNTED_MODEL,
+            client=client,
+        )
+
+    assert (exc_info.value.status_code, exc_info.value.message) == (400, problem)
+    client.post.assert_not_awaited()
+
+
 class TestBedrockCountTokensEndpoint:
     def _make_handler(self) -> BedrockCountTokensConfig:
         return BedrockCountTokensConfig()

tests/unit/llms/bedrock/count_tokens/test_bedrock_count_tokens_transformation.py

@@ -311,3 +311,37 @@ def test_count_tokens_endpoint_encodes_model_id(monkeypatch):
         endpoint
         == "https://bedrock-runtime.us-east-1.amazonaws.com/model/..%2F..%2Fmodel%2Fother%3Fx%3D1%23frag/count-tokens"
     )
+
+
+@pytest.mark.parametrize(
+    "request_data",
+    [
+        {"model": "us.anthropic.claude-opus-4-8", "messages": [{"role": "user", "content": "hi"}]},
+        {"model": "us.anthropic.claude-opus-4-8", "messages": "hi"},
+        {
+            "model": "us.anthropic.claude-opus-4-8",
+            "messages": [{"role": "user", "content": [{"type": "text", "text": "hi"}]}, {"content": "hi"}],
+        },
+    ],
+)
+def test_count_tokens_request_problem_leaves_bodies_bedrock_can_judge_to_bedrock(request_data):
+    assert BedrockCountTokensConfig().count_tokens_request_problem(request_data) is None
+
+
+def test_count_tokens_request_problem_names_tool_calls_instead_of_asking_for_content():
+    problem = BedrockCountTokensConfig().count_tokens_request_problem(
+        {
+            "model": "us.amazon.nova-pro-v1:0",
+            "messages": [
+                {"role": "user", "content": "hi"},
+                {
+                    "role": "assistant",
+                    "tool_calls": [{"id": "1", "type": "function", "function": {"name": "t", "arguments": "{}"}}],
+                },
+            ],
+        }
+    )
+
+    assert problem is not None
+    assert "tool_calls" in problem
+    assert "must have a 'content' field" not in problem

tests/unit/llms/bedrock/count_tokens/test_bedrock_token_counter.py

@@ -74,6 +74,59 @@ async def test_claude_model_bedrock_runtime_cannot_count_is_counted_on_mantle()
     assert mantle_body["system"] == "You are a terse assistant."
 
 
+@pytest.mark.asyncio
+async def test_claude_model_message_without_content_is_rejected_without_calling_runtime_or_mantle() -> None:
+    bedrock: Final = _Bedrock(runtime=(200, {"inputTokens": 1353}), mantle=(200, {"input_tokens": 1336}))
+
+    result: Final = await _counter(bedrock).count_tokens(
+        model_to_use="global.anthropic.claude-opus-4-8",
+        messages=[{"role": "user"}],
+        contents=None,
+        deployment=DEPLOYMENT,
+        request_model="claude-opus-4-8",
+    )
+
+    assert result is not None
+    assert (result.error, result.status_code, result.error_message, result.rejected_before_provider_call) == (
+        True,
+        400,
+        "Message 0 must have a 'content' field",
+        True,
+    )
+    assert bedrock.posted_hosts() == ()
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize(
+    ("message", "problem"),
+    [
+        ({"role": "user"}, "Message 0 must have a 'content' field"),
+        ({"content": "hi"}, "Message 0 must have a 'role' field"),
+    ],
+)
+async def test_mantle_counted_model_rejects_a_malformed_message_without_calling_mantle(
+    message: dict[str, str], problem: str
+) -> None:
+    bedrock: Final = _Bedrock(runtime=(200, {"inputTokens": 1353}), mantle=(200, {"input_tokens": 1336}))
+
+    result: Final = await _counter(bedrock).count_tokens(
+        model_to_use="bedrock_mantle/anthropic.claude-opus-5-5",
+        messages=[message],
+        contents=None,
+        deployment=DEPLOYMENT,
+        request_model="claude-opus-5-5",
+    )
+
+    assert result is not None
+    assert (result.error, result.status_code, result.error_message, result.rejected_before_provider_call) == (
+        True,
+        400,
+        problem,
+        True,
+    )
+    assert bedrock.posted_hosts() == ()
+
+
 @pytest.mark.asyncio
 async def test_bedrock_runtime_count_is_kept_when_it_answers() -> None:
     bedrock: Final = _Bedrock(runtime=(200, {"inputTokens": 1353}), mantle=(200, {"input_tokens": 1336}))

tests/unit/llms/bedrock/test_bedrock_common_utils.py

@@ -1663,6 +1663,21 @@ class TestGetBedrockBaseModel:
         assert get_bedrock_base_model(input_model) == expected
 
 
+def test_region_only_deployment_takes_no_traffic_from_a_valid_model_group() -> None:
+    valid_model: Final = "bedrock/us.anthropic.claude-opus-4-6-v1"
+    router: Final = litellm.Router(
+        model_list=[
+            {"model_name": "group", "litellm_params": {"model": "bedrock/us", "aws_region_name": "us-east-1"}},
+            {"model_name": "group", "litellm_params": {"model": valid_model, "aws_region_name": "us-east-1"}},
+        ],
+        ignore_invalid_deployments=True,
+    )
+
+    assert [deployment["litellm_params"]["model"] for deployment in router.get_model_list(model_name="group")] == [
+        valid_model
+    ]
+
+
 @pytest.mark.usefixtures("_vcr_outcome_gate", "setup_and_teardown")
 class TestBedrockModelInfoWrappers:
     """Tests that BedrockModelInfo methods correctly wrap standalone functions."""

tests/unit/proxy/anthropic_endpoints/test_endpoints.py

@@ -13,6 +13,7 @@ from unittest.mock import AsyncMock, MagicMock, patch
 
 import httpx
 import pytest
+import respx
 from fastapi import Request, Response
 from fastapi.responses import StreamingResponse
 from fastapi.testclient import TestClient
@@ -1222,6 +1223,49 @@ def test_count_tokens_rejected_before_counting_logs_the_400_it_answers(
     )
 
 
+def test_count_tokens_on_bedrock_answers_400_for_a_message_without_content_without_calling_bedrock(
+    monkeypatch: pytest.MonkeyPatch, httpx_transport: None, respx_mock: respx.MockRouter
+) -> None:
+    from fastapi import FastAPI
+
+    import litellm.proxy.proxy_server as proxy_server
+    from litellm.proxy._types import UserAPIKeyAuth
+    from litellm.proxy.anthropic_endpoints.endpoints import router
+    from litellm.proxy.auth.user_api_key_auth import user_api_key_auth
+
+    monkeypatch.delenv("AWS_BEARER_TOKEN_BEDROCK", raising=False)
+    monkeypatch.setattr(
+        proxy_server,
+        "llm_router",
+        Router(
+            model_list=[
+                {
+                    "model_name": "count-bedrock-claude",
+                    "litellm_params": {
+                        "model": "bedrock/us.anthropic.claude-zeta-9",
+                        "aws_region_name": "us-east-1",
+                        "aws_access_key_id": "AKIATESTACCESSKEY",
+                        "aws_secret_access_key": "test-secret",
+                    },
+                    "model_info": {"id": "count-bedrock-claude-deployment", "token_count_framing": 3},
+                }
+            ]
+        ),
+    )
+    app: Final = FastAPI()
+    app.include_router(router)
+    app.dependency_overrides[user_api_key_auth] = lambda: UserAPIKeyAuth()
+
+    response: Final = TestClient(app).post(
+        "/v1/messages/count_tokens", json={"model": "count-bedrock-claude", "messages": [{"role": "user"}]}
+    )
+
+    assert response.status_code == 400, response.text
+    assert response.json()["error"]["type"] == "invalid_request_error"
+    assert response.json()["error"]["message"] == "Message 0 must have a 'content' field"
+    assert len(respx_mock.calls) == 0
+
+
 class TestEventLoggingBatchEndpoint:
     """Test the stubbed event logging batch endpoint"""
 

tests/unit/proxy/pass_through_endpoints/test_llm_pass_through_endpoints.py

@@ -7659,8 +7659,16 @@ _BEDROCK_COUNT_ALIAS: Final = "count-log-bedrock"
 _BEDROCK_COUNT_MODEL: Final = "bedrock/us.anthropic.claude-haiku-4-5-20251001-v1:0"
 
 
+_BEDROCK_COUNT_BODY: Final = MappingProxyType(
+    {"model": _BEDROCK_COUNT_ALIAS, "messages": [{"role": "user", "content": "hi"}]}
+)
+
+
 async def _bedrock_native_count(
-    monkeypatch: pytest.MonkeyPatch, respx_mock: respx.MockRouter, aws_reply: httpx.Response
+    monkeypatch: pytest.MonkeyPatch,
+    respx_mock: respx.MockRouter,
+    aws_reply: httpx.Response,
+    request_body: Mapping[str, object] = _BEDROCK_COUNT_BODY,
 ) -> dict[str, object]:
     from litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints import (
         handle_bedrock_count_tokens,
@@ -7700,7 +7708,7 @@ async def _bedrock_native_count(
         request=request,
         fastapi_response=Response(),
         user_api_key_dict=MagicMock(),
-        request_body={"model": _BEDROCK_COUNT_ALIAS, "messages": [{"role": "user", "content": "hi"}]},
+        request_body=dict(request_body),
     )
 
 
@@ -7754,6 +7762,81 @@ async def test_bedrock_native_count_tokens_logs_the_provider_status_when_the_cou
     )
 
 
+@pytest.mark.parametrize(
+    ("request_body", "problem"),
+    [
+        ({"model": _BEDROCK_COUNT_ALIAS}, "Request must contain content to count tokens"),
+        ({"model": _BEDROCK_COUNT_ALIAS, "messages": []}, "messages parameter is required for Converse input"),
+        ({"model": _BEDROCK_COUNT_ALIAS, "messages": [{"role": "user"}]}, "Message 0 must have a 'content' field"),
+        ({"model": _BEDROCK_COUNT_ALIAS, "messages": [{"content": "hi"}]}, "Message 0 must have a 'role' field"),
+    ],
+)
+@pytest.mark.asyncio
+async def test_bedrock_native_count_tokens_answers_400_for_a_malformed_body_without_calling_bedrock(
+    monkeypatch: pytest.MonkeyPatch,
+    httpx_transport: None,
+    respx_mock: respx.MockRouter,
+    request_body: Mapping[str, object],
+    problem: str,
+) -> None:
+    with pytest.raises(HTTPException) as exc_info:
+        await _bedrock_native_count(
+            monkeypatch, respx_mock, httpx.Response(200, json={"inputTokens": 12}), request_body
+        )
+
+    assert (exc_info.value.status_code, exc_info.value.detail) == (400, {"error": problem})
+    assert len(respx_mock.calls) == 0
+
+
+@pytest.mark.parametrize(
+    ("model", "model_id"),
+    [
+        ("us", "us"),
+        ("global", "global"),
+        ("apac", "apac"),
+        ("bedrock/us", "us"),
+        ("converse/us", "us"),
+        ("invoke/us", "us"),
+        ("bedrock/converse/global", "global"),
+    ],
+)
+@pytest.mark.asyncio
+async def test_bedrock_native_count_tokens_for_a_bare_region_prefix_answers_bedrocks_own_400(
+    monkeypatch: pytest.MonkeyPatch,
+    httpx_transport: None,
+    respx_mock: respx.MockRouter,
+    model: str,
+    model_id: str,
+) -> None:
+    from litellm.proxy.pass_through_endpoints.llm_passthrough_endpoints import (
+        handle_bedrock_count_tokens,
+    )
+
+    monkeypatch.setattr("litellm.proxy.proxy_server.llm_router", None)
+    monkeypatch.delenv("AWS_BEARER_TOKEN_BEDROCK", raising=False)
+    monkeypatch.setenv("AWS_ACCESS_KEY_ID", "AKIATESTACCESSKEY")
+    monkeypatch.setenv("AWS_SECRET_ACCESS_KEY", "test-secret")
+    monkeypatch.setenv("AWS_REGION_NAME", "us-east-1")
+    invalid_model: Final = {"message": "The provided model identifier is invalid."}
+    respx_mock.post(host="bedrock-runtime.us-east-1.amazonaws.com", path=f"/model/{model_id}/count-tokens").mock(
+        return_value=httpx.Response(400, json=invalid_model)
+    )
+    request: Final = Request({"type": "http", "method": "POST", "path": f"/bedrock/model/{model}/count-tokens", "headers": []})
+
+    with pytest.raises(HTTPException) as exc_info:
+        await handle_bedrock_count_tokens(
+            endpoint=f"model/{model}/count-tokens",
+            request=request,
+            fastapi_response=Response(),
+            user_api_key_dict=MagicMock(),
+            request_body={"model": model, "messages": [{"role": "user", "content": "hi"}]},
+        )
+
+    assert exc_info.value.status_code == 400
+    assert json.loads(exc_info.value.detail["error"]) == invalid_model
+    assert len(respx_mock.calls) == 1
+
+
 @pytest.mark.asyncio
 async def test_bedrock_native_count_tokens_without_a_model_logs_the_400_it_answers(
     token_count_log_lines: Callable[[], tuple[dict[str, object], ...]],
@@ -7820,9 +7903,7 @@ async def test_bedrock_native_count_tokens_with_a_region_only_model_answers_the_
             request_body={"model": "us"},
         )
 
-    assert exc_info.value.detail == {
-        "error": "CountTokens processing error: Request must contain content to count tokens"
-    }
+    assert exc_info.value.detail == {"error": "Request must contain content to count tokens"}
     assert token_count_log_lines() == (
         {
             "litellm_call_id": "bedrock-native-region-only",

tests/unit/proxy/test_proxy_token_counter.py

@@ -1080,33 +1080,32 @@ async def test_bedrock_handler_httpx_error_status_code_propagation():
         response=mock_response,
     )
 
-    with patch.object(handler, "validate_count_tokens_request"):
-        with patch.object(handler, "_get_aws_region_name", return_value="us-west-2"):
-            with patch.object(handler, "transform_anthropic_to_bedrock_count_tokens", return_value={}):
-                with patch.object(
-                    handler,
-                    "get_bedrock_count_tokens_endpoint",
-                    return_value="https://example.com",
-                ):
-                    with patch.object(handler, "_sign_request", return_value=({}, "{}")):
-                        with patch("litellm.llms.bedrock.count_tokens.handler.get_async_httpx_client") as mock_client:
-                            mock_async_client = AsyncMock()
-                            mock_async_client.post = AsyncMock(side_effect=http_error)
-                            mock_client.return_value = mock_async_client
+    with patch.object(handler, "_get_aws_region_name", return_value="us-west-2"):
+        with patch.object(handler, "transform_anthropic_to_bedrock_count_tokens", return_value={}):
+            with patch.object(
+                handler,
+                "get_bedrock_count_tokens_endpoint",
+                return_value="https://example.com",
+            ):
+                with patch.object(handler, "_sign_request", return_value=({}, "{}")):
+                    with patch("litellm.llms.bedrock.count_tokens.handler.get_async_httpx_client") as mock_client:
+                        mock_async_client = AsyncMock()
+                        mock_async_client.post = AsyncMock(side_effect=http_error)
+                        mock_client.return_value = mock_async_client
 
-                            with pytest.raises(BedrockError) as exc_info:
-                                await handler.handle_count_tokens_request(
-                                    request_data={
-                                        "model": "test",
-                                        "messages": [{"role": "user", "content": "hello"}],
-                                    },
-                                    litellm_params={},
-                                    resolved_model="anthropic.claude-3-sonnet",
-                                )
+                        with pytest.raises(BedrockError) as exc_info:
+                            await handler.handle_count_tokens_request(
+                                request_data={
+                                    "model": "test",
+                                    "messages": [{"role": "user", "content": "hello"}],
+                                },
+                                litellm_params={},
+                                resolved_model="anthropic.claude-3-sonnet",
+                            )
 
-                            assert exc_info.value.status_code == 403
-                            # Message should be the raw response text
-                            assert exc_info.value.message == "Forbidden - Invalid credentials"
+                        assert exc_info.value.status_code == 403
+                        # Message should be the raw response text
+                        assert exc_info.value.message == "Forbidden - Invalid credentials"
 
 
 @pytest.mark.asyncio