Missing `reasoning.mode` (for `pro` mode) setting for GPT 5.6 models
Open 💬 4 comments Opened Jul 13, 2026 by Manouchehri
What variant of Codex are you using?
CLI
What feature would you like to see?
Currently, codex has no way of setting the reasoning.mode parameter to pro.
i.e. these should work (at least for those of us using API billing):
codex exec -m gpt-5.6-sol -c model_reasoning_effort=xhigh -c model_reasoning_mode=pro "Hello world"
codex exec -m gpt-5.6-luna -c model_reasoning_effort=medium -c model_reasoning_mode=pro "Hello world"https://developers.openai.com/api/docs/guides/reasoning#reasoning-mode
Related to (but not a duplicate of, as these issues are not talking about the reasoning _mode_, but rather the older GPT 5.5 Pro model, which does not use reasoning.mode): #27077 and #28709
Additional information
This is a pretty simple fix. (Not going to waste my time doing a PR, since I know OpenAI's current policy is to refuse external PRs.)
4 Comments
The current workaround I have in production systems, is having my LLM gateway offer
gpt-5.6-sol-pro/gpt-5.6-terra-pro/gpt-5.6-luna-proas virtual models, so codex doesn't have to know aboutreasoning.mode.Not an ideal fix though, and does not really help other people.
which LLM Gateway do you use to provide the virtual model please? Just struggled with this also.. thx!
Thank you for mentioning the gateway workaround. It works. Based on your experience, which model is more powerful: the chatgpt.com Sol Pro model or the API Sol Pro Max model (mode: pro, effort: max)? OpenAI doesn't disclose what reasoning effort the chatgpt.com Sol Pro model utilizes.
It's a closed source internal thing. Not terribly hard for other LLM gateways to do though.