Max tokens
Also known as: max_tokens · output limit
Max tokens (often max_tokens or max_completion_tokens) caps how many tokens the model may generate in a single response.
- Prevents endless, looping, or runaway output.
- Controls cost (you pay per generated token).
- Set it based on the task: a one-word label needs ~10; an essay needs thousands.
Distinct from context window: max-tokens limits the *output*; the context window limits *input + output together*. Don't mix them up.