[Bug]: tools.3.custom.input_examples: Extra inputs are not permitted sporadic error with Claude Code

[IceKickr]

What happened?

The following error is seen sporadically, on and off - with Claude Code - this started happening today.

=->claude -p "Say hello"
API Error: 400 {"error":{"message":"{"message":"tools.3.custom.input_examples: Extra inputs are not permitted"}. Received Model Group=claude-sonnet-4-5-20250929\nAvailable Model Group Fallbacks=None","type":"None","param":"None","code":"400"}}

Claude Code version hasn't changed.

Relevant log output

API Error: 400 {"error":{"message":"{\"message\":\"tools.3.custom.input_examples: Extra inputs are not permitted\"}. Received Model Group=claude-sonnet-4-5-20250929\nAvailable Model Group Fallbacks=None","type":"None","param":"None","code":"400"}}

LiteLLM proxy config:

  - model_name: claude-sonnet-4-5-20250929
    litellm_params:
      model: bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.79.1

Twitter / LinkedIn details

No response

[ying80]

Can verify, started to see this error this morning (PST)

[IceKickr]

Can verify, started to see this error this morning (PST)

Are you still seeing the issue? It just magically went away for me. Which makes me think that Amazon rolled out an update with an issue, then rolled it back...

[ying80]

Can verify, started to see this error this morning (PST)

Are you still seeing the issue? It just magically went away for me. Which makes me think that Amazon rolled out an update with an issue, then rolled it back...

I did some tests, it seem to be random. But cc version 36 is the most stable.

[mzagar]

Looks like Claude Code v2.0.42 introduced a regression (sometime after v2.0.36) where it sends input_examples in tool definitions, causing failures with AWS Bedrock. This issue doesn't occur in Claude Code v2.0.36. See: anthropics/claude-code#11678

I tested whether LiteLLM's drop_params: true configuration could work around this issue. Results show that drop_params appears to only work for top-level parameters, not nested fields within tool definitions.

Test Setup

• LiteLLM Version: main-latest (Docker: ghcr.io/berriai/litellm:main-latest)
• Model: bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0
• Configuration (test-config.yaml):

model_list:
  - model_name: bedrock-claude-4.5-sonnet
    litellm_params:
      model: bedrock/us.anthropic.claude-sonnet-4-5-20250929-v1:0
      aws_access_key_id: os.environ/AWS_ACCESS_KEY_ID
      aws_secret_access_key: os.environ/AWS_SECRET_ACCESS_KEY
      aws_region_name: us-east-1

litellm_settings:
  drop_params: true

Docker command:

docker run -p 8000:4000 \
  -e AWS_ACCESS_KEY_ID="$AWS_ACCESS_KEY_ID" \
  -e AWS_SECRET_ACCESS_KEY="$AWS_SECRET_ACCESS_KEY" \
  -v $(pwd)/test-config.yaml:/app/config.yaml \
  ghcr.io/berriai/litellm:main-latest \
  --config /app/config.yaml

Test Results

Request WITHOUT input_examples - Works

curl "http://localhost:8000/v1/messages?beta=true" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -H "x-api-key: anything" \
  -d '{
    "model": "bedrock-claude-4.5-sonnet",
    "max_tokens": 1024,
    "tools": [
      {
        "name": "grep_tool",
        "description": "Search for patterns",
        "input_schema": {
          "type": "object",
          "properties": {"pattern": {"type": "string"}},
          "required": ["pattern"]
        }
      }
    ],
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Response:

{
  "model": "claude-sonnet-4-5-20250929",
  "id": "msg_bdrk_01JtkTCNc6TDB72duLLTfuSn",
  "type": "message",
  "role": "assistant",
  "content": [
    {
      "type": "text",
      "text": "Hello! How can I help you today?"
    }
  ],
  "stop_reason": "end_turn",
  "stop_sequence": null,
  "usage": {
    "input_tokens": 555,
    "cache_creation_input_tokens": 0,
    "cache_read_input_tokens": 0,
    "output_tokens": 12
  }
}

Request WITH input_examples - Fails

curl "http://localhost:8000/v1/messages?beta=true" \
  -H "Content-Type: application/json" \
  -H "anthropic-version: 2023-06-01" \
  -H "x-api-key: anything" \
  -d '{
    "model": "bedrock-claude-4.5-sonnet",
    "max_tokens": 1024,
    "tools": [
      {
        "name": "grep_tool",
        "description": "Search for patterns",
        "input_schema": {
          "type": "object",
          "properties": {"pattern": {"type": "string"}},
          "required": ["pattern"]
        },
        "input_examples": [{"pattern": "search term"}]
      }
    ],
    "messages": [{"role": "user", "content": "Hello"}]
  }'

Response:

{
  "error": {
    "message": "{\"message\":\"tools.0.custom.input_examples: Extra inputs are not permitted\"}. Received Model Group=bedrock-claude-4.5-sonnet\nAvailable Model Group Fallbacks=None",
    "type": "None",
    "param": "None",
    "code": "400"
  }
}

Stack Trace

09:00:59 - LiteLLM Proxy:ERROR: endpoints.py:214 - litellm.proxy.proxy_server.anthropic_response(): Exception occured - {"message":"tools.0.custom.input_examples: Extra inputs are not permitted"}
Traceback (most recent call last):
  File "/usr/lib/python3.13/site-packages/litellm/llms/custom_httpx/llm_http_handler.py", line 1875, in async_anthropic_messages_handler
    response = await async_httpx_client.post(
  File "/usr/lib/python3.13/site-packages/litellm/litellm_core_utils/logging_utils.py", line 190, in async_wrapper
    result = await func(*args, **kwargs)
  File "/usr/lib/python3.13/site-packages/litellm/llms/custom_httpx/http_handler.py", line 403, in post
    raise e
  File "/usr/lib/python3.13/site-packages/litellm/llms/custom_httpx/http_handler.py", line 359, in post
    response.raise_for_status()
  File "/usr/lib/python3.13/site-packages/httpx/_models.py", line 829, in raise_for_status
    raise HTTPStatusError(message, request=request, response=self)
httpx.HTTPStatusError: Client error '400 Bad Request' for url 'https://bedrock-runtime.us-east-1.amazonaws.com/model/us.anthropic.claude-sonnet-4-5-20250929-v1:0/invoke'

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.13/site-packages/litellm/proxy/anthropic_endpoints/endpoints.py", line 153, in anthropic_response
    responses = await llm_responses
  File "/usr/lib/python3.13/site-packages/litellm/router.py", line 2895, in _ageneric_api_call_with_fallbacks
    raise e
  File "/usr/lib/python3.13/site-packages/litellm/router.py", line 2882, in _ageneric_api_call_with_fallbacks
    response = await self.async_function_with_fallbacks(**kwargs)
  File "/usr/lib/python3.13/site-packages/litellm/router.py", line 4311, in async_function_with_retries
    response = await self.make_call(original_function, *args, **kwargs)
  File "/usr/lib/python3.13/site-packages/litellm/llms/custom_httpx/llm_http_handler.py", line 1884, in async_anthropic_messages_handler
    raise self._handle_error(
  File "/usr/lib/python3.13/site-packages/litellm/llms/custom_httpx/llm_http_handler.py", line 3582, in _handle_error
    raise provider_config.get_error_class(
litellm.llms.base_llm.chat.transformation.BaseLLMException: {"message":"tools.0.custom.input_examples: Extra inputs are not permitted"}
INFO:     172.17.0.1:55798 - "POST /v1/messages?beta=true HTTP/1.1" 400 Bad Request

Findings

The drop_params: true configuration does not strip nested fields like input_examples within tool definitions. It appears to only work for top-level request parameters.

While this looks to be a Claude Code issue (not honoring the OpenAI API spec), if additional_drop_params could be enhanced to support nested tool fields (e.g., tools[].input_examples), it would provide a useful workaround for this type of client-side issue.

Current workaround: Downgrade to Claude Code v2.0.36, which doesn't include input_examples in tool definitions.

[IceKickr]

@mzagar What's very confusing to me is the fact that this started happening for me last Friday, and the completely went away on Saturday - using the same version of Claude Code (v2.0.42).

[mzagar]

@IceKickr yes weird. but even with this workaround i tried now stopped getting tools[].input_examples error, instead now i get antrhopic_beta header error mentioned in above claude code bug issue so not much progress (i'm using litellm + sonnet 4.5 on aws bedrock) 🤷‍♂️

at least cc 2.0.36 seems to be working without issues...

[LeonardMH]

See this related issue in the claude-code tracker: anthropics/claude-code#11678

This seems to be the result of some A/B testing, hence the randomness on whether it works or not. Removing the ~/.claude/statsig folder every time you start claude reliably allows me to use 2.0.42 + Sonnet 4.5 through LiteLLM.

I have just set the following command alias for now alias blaude = rm -rf ~/.claude/statsig && claude

EDIT: I didn't see this issue had already been mentioned, but the solution is still helpful I think so I'll leave the comment. Downgrading to 2.0.36 wasn't working for me.

[mzagar]

thanks @LeonardMH!
tried it and 2.0.42 works for my setup.
also mentioned in comments here: #16718

[bocytko]

I get

API Error: 400 {"error":{"message":"{\"message\":\"context_management.edits.0: Input tag 'clear_thinking_20251015' found using 'type' does not match any of the expected tags: 'clear_tool_uses_20250919'\"}

for Haiku 4.5 and Sonnet 4 with enabling thinking. The workaround mentioned by @LeonardMH works reliably for me as well.