Ability to enable and disable thinking for compatible models

howie

Claude 3.7 and other thinking enabled models are great! But I don't always want a thinking enabled model to go through those steps for the same reason I don't always want Kagi Search to be invoked on Assistant queries - it saves on usage quota, and it saves on time.

Consider adding a switch next to the "Web search" switch for thinking-enabled AI models, to allow us to disable the thinking step, or enable it if needed.

gladiator2339

That's simply not possible AFAIK. Thinking(reasoning) is how those models work and the decision happens inside the model and is not toggleable. You can always switch model mid-chat though.

Luis

As @gladiator2339 mentioned, selecting a reasoning model enables its reasoning capabilities, which cannot be disabled. For KI (multi-step Assistant) it can decide to use reasoning or not.

From our perspective, the ideal user experience is for the Assistant (currently KI) to automatically use reasoning when appropriate for the task, without needing user intervention.

I hope this helps addressing your question. Pls let us know.

howie

Luis Claude 3.7 offers "extended thinking" as a toggle.

Maybe these are two separate models. It looks as if the "thinking" flag being passed to the API changes this behavior.

curl https://api.anthropic.com/v1/messages \
     --header "x-api-key: $ANTHROPIC_API_KEY" \
     --header "anthropic-version: 2023-06-01" \
     --header "content-type: application/json" \
     --data \
'{
    "model": "claude-3-7-sonnet-20250219",
    "max_tokens": 20000,
    "thinking": {
        "type": "enabled",
        "budget_tokens": 16000
    },
    "messages": [
        {
            "role": "user",
            "content": "Are there an infinite number of prime numbers such that n mod 4 == 3?"
        }
    ]
}'

RoxyRoxyRoxy

howie There's a discussion about how to implement Extended Thinking here: #6365