Why are there three different `gpt-5-codex` variants?
Resolved 💬 5 comments Opened Sep 18, 2025 by PrayagS Closed Sep 30, 2025
💡 Likely answer: A maintainer (dedrisian-oai, contributor)
responded on this thread — see the highlighted reply below.
<img width="400" height="247" alt="Image" src="https://github.com/user-attachments/assets/3117ded8-8956-49c2-aa52-8aa8b7021bee" />
Today when I fire up codex CLI, I suddenly see three different variants which are controlling the reasoning effort (saw in /status).
I remember the announcement post saying,
GPT‑5-Codex adapts how much time it spends thinking more dynamically based on the complexity of the task.
If I understand it correctly, they're talking about time spent thinking which is essentially the same as reasoning effort in my book. If it was supposed to figure it out by itself, why do I have three options? Is this even intended?
I'm on version 0.38.0
5 Comments
...because that control is still relevant
What your'e trying to say is: "where's auto?" I suppose.
Sure. And also trying to know if there's a source that confirms whether the
high,low, etc. in these Codex variants denote the upper cap of the thinking budget.As in, it is still possible for the gpt-5-codex-high model to finish quickly with minimal thinking if it thinks the task is very trivial.
At least in Opus and other thinking models, I have seen that they can spend more time thinking just because they can not because they need to.
"budget" suggests upper-limit I guess but maybe its just me;
I'd like to know an answer to this too.
Would recommend just using
gpt-5-codex mediumas it does a good job finding the right amount to think.