Length, stops & cost controls

max_tokens / max_completion_tokens

• max_tokens limits the number of generated tokens and is supported by Azure endpoints.
• OpenAI’s reference also documents max_completion_tokens as the preferred modern parameter, and notes max_tokens may be deprecated / incompatible with some reasoning models.

Use these to:

• bound cost
• bound latency
• enforce concise answers

stop

• Up to a small number of stop sequences can end generation.
• Useful for code blocks, list truncation control, or multi-part outputs.

Useful executive takeaway

Length limits are one of the strongest cost controls you have, independent of decoding randomness.

Next: Determinism & Debugging.