Clarified how max_output_tokens interacts with thinking on /gemini-api/docs/thinking and /gemini-api/docs/generate-content/thinking: it caps the combined total of thinking tokens and visible output tokens as an infrastructure cutoff (wit…
Changelog
Thinking
Clarified how max_output_tokens interacts with thinking on /gemini-api/docs/thinking and /gemini-api/docs/generate-content/thinking: it caps the combined total of thinking tokens and visible output tokens as an infrastructure cutoff (without changing thinking budget allocation). Hitting the limit during reasoning can stop generation with truncated or empty output while still billing for thinking tokens generated; lower thinking_level instead of a small max_output_tokens to reduce cost or latency without truncating responses.
