noticeable regression in model reliability and speed over the past few days

Resolved 💬 2 comments Opened May 6, 2026 by federicotorrisi Closed May 6, 2026
💡 Likely answer: A maintainer (github-actions[bot], contributor) responded on this thread — see the highlighted reply below.

What version of the Codex App are you using (From “About Codex” dialog)?

codex-cli 0.128.0

What subscription do you have?

pro (200$)

What platform is your computer?

Linux 7.0.3-zen1-2-zen x86_64

What issue are you seeing?

over the past few days, all codex models i use - 5.5, 5.4, and 5.4 mini - seem to have become significantly less reliable.
since the release of 5.5, 5.4 has felt noticeably worse, but now 5.5 itself is also unreliable, and 5.4 mini has become both unreliable and much slower.
the issue is not limited to complex tasks. the models now frequently fail even at very basic tasks: sometimes they ignore explicit instructions, sometimes make unnecessary changes or do things that were never requested.
nothing has changed on my side. i haven’t changed my prompting style, usage habits, repository setup, skills, instructions, or workflow.

with comparable prompts, the output quality is simply worse than it was a few days ago.
it feels as if the models are being routed differently, quantized, or otherwise degraded.

What steps can reproduce the bug?

use codex cli

What is the expected behavior?

_No response_

Additional information

_No response_

View original on GitHub ↗

2 Comments

github-actions[bot] contributor · 3 months ago

Potential duplicates detected. Please review them and close your issue if it is a duplicate.

  • #20589
  • #20988

Powered by Codex Action

etraut-openai contributor · 3 months ago

I'm not aware of any changes that would explain a degradation like the one you're describing here.

If you see more of this, please use /feedback to upload the session details.