Symptoms

• The response ends early or reports a length finish reason.
• Reasoning or media generation consumes budget before visible output appears.

Likely causes

• max_tokens was copied from a tiny smoke test.
• The task needs more output or internal reasoning budget than expected.

How to diagnose

• Inspect finish reason and token usage when available.
• Retry one controlled request with a larger output allowance.
• Check whether the endpoint uses max_tokens or another protocol-specific field.

Fixes

• Raise the output limit within the model and account constraints.
• Ask for a shorter answer or split the work into stages.
• Avoid unbounded maximums in public or automated workloads.

How to verify

The response completes with the expected finish reason and content.

Resources

• Base URL and authentication
• FAQ and troubleshooting