Extra high is not taking it’s time to think / explore code

Resolved 💬 4 comments Opened Nov 21, 2025 by guidedways Closed Nov 21, 2025

💡 Likely answer: A maintainer (guidedways, contributor) responded on this thread — see the highlighted reply below.

What version of Codex is running?

0.61.0

What subscription do you have?

Pro

Which model were you using?

gpt codex max extra high

What platform is your computer?

macOS

What issue are you seeing?

I’ve been using Codex Max Extra High primarily (given the work I do with it requires more accurate responses, through examination of code and so on).

Unfortunately so far I feel:

It isn’t slow in thinking, quite the opposite. It responds quickly and mostly incorrectly
When it begins to fix something, it mostly gets it wrong every single time during the first few attempts and needs to be coaxed into the _correct_ approach by giving it additional context and code to explore, which it should have on its own based on (a thorough) initial prompt.

It’s almost giving me vibes of the model that should not be named (ok: Claude sonnet), trigger happy to try and fix everything incorrectly.

It no longer feels like the solid, slow moving but thoroughly examining Codex 5.1 High we know and love.

What steps can reproduce the bug?

Really anything - ask it to find the reason why an app is crashing and give it crash logs, it’ll mostly get it wrong. Then if asked to fix something in a certain way, it won’t infer what I mean and will get it wrong again.

Unless I tell it to do A then B then C in exact steps, it gets it wrong every time. Almost doesn’t think through as if it’s rushing to a conclusion.

What is the expected behavior?

I felt Extra High would mean /review style static analysis where the model really goes on a deep dive to come back with an accurate answer. /review but without the review - actual bug hunting / code smells sniffing, beast of a coder.

Instead it feels like Sonnet++ / Opus++ right now (and they’re unusable for the same reasons).

Additional information

_No response_

View original on GitHub ↗

4 Comments

aurindo10 · 9 months ago

I'm having the same problem

guidedways contributor · 9 months ago

I've switched to gpt-5.1-codex high to compare, and I think with this setting codex feels consistent, grounded and closer to the accuracy we've come to expect from Codex.

I'm convinced that _faster_ != _smarter_ when it comes to these models. codex-max ex-high is faster, but feels like it's skimming over the finer details.

Shouldn't Codex Max Extra High > Codex 5.1 High ?

etraut-openai contributor · 9 months ago

Thanks for the bug report. We'll share the feedback with the team responsible for training our codex models. If you see other model behavior that you'd like to report, please use the /feedback command in the CLI. That will provide us with additional details including logs and conversation text.

Abdulnaser97 · 9 months ago

having the same problem too Codex 5.1 High > Codex Max Extra High