Reasoning Effort

What is a Reasoning Effort

Reasoning effort is a setting that controls how much a model deliberates internally before producing its final answer. Lower effort favors a fast, more direct response, useful for simple or well-understood tasks. Higher effort lets the model work through more of the problem before committing to an answer, which tends to help on genuinely hard or ambiguous questions.

This is a real tradeoff, not a free upgrade: more reasoning effort costs more time and typically more tokens for the same question, and on a task that is genuinely simple, extra deliberation mostly adds latency without adding accuracy. Matching effort level to what the task actually needs is a meaningful lever for both speed and cost, not just a dial to always max out.

Reasoning effort is distinct from temperature, temperature controls randomness in what gets said, reasoning effort controls how much internal work goes into deciding what to say. It is also distinct from output length, a low-effort answer can still be long, and a high-effort one can still be a short, carefully chosen response.