Thinking Budget

A user- or developer-controllable limit on how many reasoning tokens a Reasoning Model is allowed to spend on a given query before it must produce a final answer, used to trade off answer quality against latency and cost.