ChatRequest.think=true does not enable model thinking

[changweige]

JS ollama SDK version is 0.5.16.

ChatRequest.think=true does not enable model deep thinking. My workaroud is as follows:

      // Create streaming chat completion
      const response = await client.chat({
        model: modelName,
        messages: messages,
        stream: true,
        think: deepThinkingEnabled ? undefined : false, // XXX: `true` is enabling deep thinking mode in Ollama, is it a bug?
        options: {
          temperature: modelParams.temperature,
          top_p: modelParams.topP,
        },
      });

However， I think it's a bug?

[drifkin]

which model are you seeing problems with? it requires model support and also updated templates (so also be sure to ollama pull any older models)

[changweige]

@drifkin I was running Qwen3-30b

[wadealexc]

Running qwen3:4b, i can confirm that the think parameter in ollama.chat does not behave how i would expect.

For this model, I noticed:

• not supplying a think parameter at all still caused the model to think; output was prefaced with a ... section
• think: false had the same outcome
• think: true caused the model to hang for over a minute before responding. there was no ... section in its response, but it took about the same amount of time as if the think section had been there.