Hello! I'm trying to enable thinking mode for Claude.

Workaround

Right now, I have to proxy out to another endpoint where I inject the right parameters into the requests. Safe to say this is a bit much for a parameter change 😅

Ideal Solution

I'd love a solution like this:

llm=anthropic.LLM(
    model="claude-3-7-sonnet-20250219",
    temperature=1,
    thinking={
        "enabled": True,
        "max_tokens": 4096,
    },
)

Currently there's no parameter for thinking in the anthropic LLM initializer.

By the way, I've looked into fixing this with a draft PR (#1721), but frankly don't have the context when contributing to feel confident in making a submission. If someone wants to point me in the right direction, I'd be happy to write the code.

Thanks!