[Bug]: Reasoning Content is not included in response from bedrock converse for sonnet 3.7

[talalryz]

What happened?

Making the curl request:

curl ... --data '{
  "model": "sonnet-3.7",
  "max_tokens": 128000,
  "thinking": {
        "type": "enabled",
        "budget_tokens": 3200
    },
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "what is 1+2?"
    }
  ]
}'

the response is:

{"id":"chatcmpl-99a791ff-b69e-4cac-83ca-1f3726fd934b","created":1740441961,"model":"us.anthropic.claude-3-7-sonnet-20250219-v1:0","object":"chat.completion","system_fingerprint":null,"choices":[{"finish_reason":"stop","index":0,"message":{"content":"The sum of 1 and 2 is 3.","role":"assistant","tool_calls":null,"function_call":null}}],"usage":{"completion_tokens":51,"prompt_tokens":49,"total_tokens":100,"completion_tokens_details":null,"prompt_tokens_details":null}}

and does have a thinking or reasoning block e.g., I would expected to have had something like this in the response:

{
            "type": "thinking",
            "thinking": " ..",
},

My litellm config looks like:

model_list:
  - model_name: sonnet-3.7
    litellm_params:
      model: bedrock/converse/us.anthropic.claude-3-7-sonnet-20250219-v1:0

Are you a ML Ops Team?

No

What LiteLLM version are you on ?

v1.61.15-nightly

Twitter / LinkedIn details

No response

[krrishdholakia]

Hey @talalryz adding this for the Anthropic API now - #8778

Will work on bedrock support next

[talalryz]

Thank you!! I'll close the issue