How to set reasoning effort?
Resolved 💬 7 comments Opened Apr 16, 2025 by zakkor Closed Apr 29, 2025
💡 Likely answer: A maintainer (tibo-openai, collaborator)
responded on this thread — see the highlighted reply below.
low/medium/high
7 Comments
These cannot be set right now, but will accept a PR. This should be conditional on the model as not all models accept a reasoning effort.
What's the default?
Option A — CLI flag/env (e.g. --reasoning-effort or REASONING_EFFORT)
• Chosen once at launch → fixed for that run.
• Zero interactive code changes; works in CI/headless.
Option B — Interactive prompt in /model command (my preference)
• When any “o*” model is selected, prompt: low / medium / high.
• Users can change effort any time by re‑running /model (even with same model).
• Keeps startup simple; dynamic and intuitive for interactive use.
@tibo-openai Which approach do you prefer ?
Let me know and I’ll implement accordingly.
Hii @tibo-openai
I've implemented the
--reasoningCLI flag to allow users to customize the reasoning effort level used by the OpenAI 'o' models.Please review PR #314
doesn't work for me
<img width="501" height="118" alt="Image" src="https://github.com/user-attachments/assets/97c1f0d6-0065-4ad2-b263-379d945b0101" />
codex --config model_reasoning_effort="high"is the term extra high and max are just marketing terms in chatgpt web interface? or it also applies to codex cli?