Claude 3.7 Reasoning Content via Bedrock and LiteLLM Proxy

[tseruga]

Read through the changes made in #8843 but it's not immediately obvious to me whether or not reasoning content is expected to work via the proxy server for calls made to Claude 3.7 to Bedrock with thinking enabled.

model = ChatOpenAI(
    openai_api_base="http://localhost:4000",
    model_name="Bedrock Claude 3.7",
    temperature=1.0,
    openai_api_key="xxx",
    extra_body={
        "thinking": {"type": "enabled", "budget_tokens": 1024}
    }
)

It appears that the thinking parameter is respected, but the responses from the proxy server never contain the reasoning/thinking content. provider_specific_fields is also None which I know was discussed as a location where these reasoning blocks might appear while a generalized solution was worked on.

I know this model is brand new, but just wanted to make sure I understood where the effort currently stood - and that I'm not wildly misunderstanding things.

[tseruga]

On further investigation, as far as I can tell LiteLLM is behaving as expected

[krrishdholakia]

Hey @tseruga were you able to resolve your issue?