Disabling or reducing <think>...</think> content

General
1

Hello, everyone.

The question is, is it possible to somehow influence the length of the content returned in think tag, or disable it altogether and not return it in order to get the result of other content faster?

I tried to use this through a prompt, but I don’t get any effect. It seems that prompts are not always taken into account when generating the answer.

        const options = {
      search_mode: searchMode,
      ...(searchAfterDate && { search_after_date_filter: searchAfterDate }),
      ...(searchBeforeDate && { search_before_date_filter: searchBeforeDate }),
      temperature: 0.7,
      response_format: {
        type: 'json_schema',
        json_schema: {
          schema: configModel.schema,
        },
      },
    };

    const result = streamText({
      model: myPerplexity('sonar-deep-research'),
      messages: [
        {
          role: 'system',
          content: configModel.prompt,
        },
        ...convertToModelMessages(messages),
      ],
      providerOptions: {
        perplexity: options,
      },
    });