Reasoning

Reasoning is off by default for speed. Turn it on per request:

Endpoint	Field	Values
Chat Completions	reasoning_effort	none, low, medium, high
Responses	reasoning.effort	none, low, medium, high

{ "model": "large-v2", "reasoning_effort": "medium", "messages": [{ "role": "user", "content": "Prove there are infinitely many primes." }] }

Reasoning tokens are billed as output and reported in usage.completion_tokens_details.reasoning_tokens (Chat) or usage.output_tokens_details.reasoning_tokens (Responses).

With reasoning on, the output limit is at least 256 tokens so the answer is not starved by the thinking. large-v2 always reasons.

The Messages endpoint does not accept thinking.