[Bug] Opus 4.8 Model Quality Degradation

Status Open
Reported on v2.1.183
Maintainer reply None cached
Activity 5 comments · opened Jun 19, 2026

Bug Description
The quality of the Opus 4.8 Ultracode model today has been absolutely terrible. I don't understand how you downgraded it so much... without warning. Today it looks like a model from a year ago... you can't do this.

Environment Info

  • Platform: darwin
  • Terminal: iTerm.app
  • Version: 2.1.183
  • Feedback ID: 41ca8cc4-0574-4471-8688-ac359270a990

Errors

[]

View original on GitHub ↗

5 Comments

gaandurian · 2 months ago

I second this. today Opus 4.8 at high effort is making terrible terrible mistakes.
tasks that are usually trivial now seem impossible. it can't even describe the difference between two presented UI styles that are very visually distinct. it's so frustrating...

gincol · 2 months ago

I have a hypothesis I'd like to put forward in case it makes sense. If it does, and you can confirm it (which I doubt), great. Otherwise, it's just a hypothesis.

For development tasks—and I mean development only, nothing to do with OpenClaw or anything like it, I repeat, only development—I've been linking for over 24 consecutive hours in 3 or 4 concurrent active sessions for several different projects.

The hypothesis: Do you penalize intensive account usage because it's not profitable? Do you lower the effort level in the model when you detect that intensive usage?

Detail: I understand there's an individual component to the loss of quality because if it were global, it would be unsustainable.

latticedefect · 2 months ago

Yeah its terrible

kwhat · 2 months ago

I am experiencing the same issues. Tasks that were solved easily a few weeks ago are riddled with mistakes. Even explicitly telling Claude about the mistakes it made and to search for other instances of that mistake results in more mistakes. I am getting increasingly frustrated with whatever has changed on your end.

CaliViking · 1 month ago

Opus 4.8 has become very verbose and is making mistakes that it did not do before. It is ignoring clear instructions.