[Roadmap] Add a toggle-able thinking mode for models/providers that allow for it

[tranhoangnguyen03]

Why
More and more models/providers are offering hybrid LLMs: regular fast response & thinking mode response.

Description
One example is Anthropic Claude API: https://docs.anthropic.com/en/docs/build-with-claude/extended-thinking
Here, you simply add a thinking param in the API request to enable the thinking mode:

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-3-7-sonnet-20250219",
    max_tokens=20000,
    
    ###### thinking params ######
    thinking={
        "type": "enabled",
        "budget_tokens": 16000
    },
    #########################

    messages=[{
        "role": "user",
        "content": "Are there an infinite number of prime numbers such that n mod 4 == 3?"
    }]
)

print(response)

The corresponding content will contain the thoughts in addition to the regular response

{
    "content": [
        {
            "type": "thinking",
            "thinking": "To approach this, let's think about what we know about prime numbers...",
            "signature": "zbbJhbGciOiJFU8zI1NiIsImtakcjsu38219c0.eyJoYXNoIjoiYWJjMTIzIiwiaWFxxxjoxNjE0NTM0NTY3fQ...."
        },
        {
            "type": "text",
            "text": "Yes, there are infinitely many prime numbers such that..."
        }
    ]
}

Requirements

• Need to compile a list of providers (including self-hosted options such as VLLM, Llama.cpp, and Ollama,...) that provide this capability and how each of them handles the request and response.
• Need to be able to detect which model offers this mode (Sonnet 3.5 does not, but Sonnet 3.7 does)
• Need some changes to the UI to enable this thinking mode for appropriate models and to specify thinking budget
• Some dependencies with the cost calculations as well

[enricoros]

@tranhoangnguyen03 I've completed this support. Which Big-AGI do you use? wbesite, docker, ?

[tranhoangnguyen03]

I think I'm on an old docker build. Let me update my deployment. Thanks!

[enricoros]

@tranhoangnguyen03 Please let me know if this works as expected, and whether the support could be extended in other directions.