Reasoning models

Use models that spend additional tokens internally on maths, planning, debugging and complex decisions.

const response = await client.chat.completions.create({
  model: "YOUR_REASONING_MODEL",
  messages: [{ role: "user", content: "Solve this step by step." }],
  max_tokens: 8192,
});
console.log(response.choices[0].message.content);
// Optional reasoning_content is returned separately when available.

Reasoning output

The final answer remains in content. When a route exposes it, reasoning_content is returned separately and may be shown in a collapsed detail.

Budget

Reasoning tokens count toward completion usage. Use at least 2048 tokens for short tasks and a larger ceiling for complex code or multi-step work.

Choose deliberately

Reasoning models are slower and consume more tokens. Prefer a regular model for simple formatting, translation and short lookups.

Need help?

Keep the request ID and contact Torve support without sharing your API key or private data.

Contact support