Not having internal COT <Think> in output of reasoning models

General
1

I’m new to using the Perplexity API but I’m trying to get a structured output response from my query and it feels really wasteful to generate all these thinking output tokens if I only care about the structured JSON output.

I do want the improved answer that comes from the reasoning version of the models. I just don’t care to see or pay for thousands, sometimes 10s of thousands of output tokens that are the model’s internal thoughts.

Is there a way to suppress this from the output and still get reasoning or if I want to use the reasoning models or do I just need to accept I’m getting a bunch of tokens I don’t need/want

I just spent like 20 cents on a single request that only provided me a bunch of thinking tokens and hit the my max token limit without providing me my desired answer

2

No, you cannot suppress thinking tokens while using Perplexity’s reasoning models. The response format parameter doesn’t remove them. Your only options:Use non-reasoning models (sonar or sonar-pro) for structured outputs without thinking tokens Stick with reasoning models and manually parse out the JSON after the section.
You must choose between reasoning quality with expensive thinking tokens, or standard models without them.