OpenAI API: tune reasoning effort and verbosity separately
reasoning_effort controls how hard the model thinks; verbosity controls how long the answer is. Adjust them instead of rewriting prompts.
OpenAI’s GPT-5 API exposes two separate knobs that people often try to control with prompt wording:
reasoning_effort: how much the model thinks and how willing it is to call tools. Default ismedium. Useminimal/lowfor extraction, formatting or classification where latency matters; go higher for complex multi-step work.verbosity: the length of the final answer (lowfor terse UX), independent of how much it thinks.
Tips from the cookbook
- Keep prompts stable and change the parameter instead of rewriting “be brief” into your prompt.
- Avoid
minimalfor multi-step planning or tool-heavy workflows. - For agentic flows, split distinct tasks across separate agent turns.