[Bug] [URGENT] Claude Opus 4.8 reasoning degradation, speed and performance regression

Description

**Bug Description** Extremely poor reasoning, even on Max effort. The model (Opus 4.8) is severely downgraded and degraded. I have been compiling relevant proof and will be consulting to pursue an action as a EU customer for what I consider to be deceptive business practices and ridiculing the user. The model can't hold multiple facts together anymore, fails to read references prior to answering even when directly nudged, is lazy and shallow. This is a behaviour that began roughly 3 days ago and this was not the behavior at the time of me purchasing x20 £200 Claude Max subscription to use it. It feels like Haiku is a more powerful model than the one you are deceptively giving me as Opus 4.8. I demand immediate fix for this and stopping this malpractice. The copy of the this will be sent to your legal team and, if no response is give, to my attorney. For past 3 days I've been trying to get your attention to this issue, and others: Claude Opus 4.8 is EXCEPTIONALLY slow, as also confirmed by multiple similar issues on Github. EDIT 23 June 2026: The degradation comes and goes. One day it's my Claude, the best in the world. The next morning - it can't tell me what folder it's working on. It's not just an "increased error rate", it's a 100 IQ points difference. Please stop doing that and stop thinking yoou can get away with that: I have been paying £200 for this subscription, not for the nerfed, for the REAL one. Leave us alone, let us enjoy using what we paid for! If you can't scale well - don't scale, we are not to blame! We chose YOU. Stop making us regret it! **Environment Info** - Platform: darwin - Terminal: ghostty - Version: 2.1.177 - Feedback ID: 1a153ce3-db91-4295-99d0-269b953917c2 **Errors** ```json [] ```