Reference · Glossary
Max tokens
An API setting that caps **how long the model's reply** can be — measured in tokens, not words.
When to use
Control cost, prevent rambling answers, or enforce concise formats ("max 150 tokens").
When not to
Setting max tokens too low cuts off mid-sentence. Leave headroom for complete answers.
Example
{ "max_tokens": 256 }Rough guide: 100 tokens ≈ 75 words of English prose.