[Bug] Anthropic API: Claude ~~3.5 Sonnet~~ 4.5 Opus performance degradation and reasoning errors

Bug Description
Opus 4.5 performance has degraded extremely in the last hours. I'm sure this is A/B testing or a bug in model selection, people will blame my setup/prompt-engineering/context-engineering/any-proompter-buzzword but I'm a very heavy user and I'm 100% positive about this being a very bad regression. It is both slower and dumber. Not even simple tasks can be accomplished now, just as an example, it just tried to:

⏺ Bash(pkill -f "next dev" 2>/dev/null; sleep 1
      cd /tmp/git-heatmap-app && rm -rf .next && pnpm dev &
      sleep 4 && open http://localhost:3000)
  ⎿  Interrupted · What should Claude do instead?

After 2 code edits in nextjs, everyone knows next has hot reload, what the hell is going on.

Environment Info

• Platform: darwin
• Terminal: ghostty
• Version: 2.0.70
• Feedback ID: 8c4005b1-e856-4220-8c63-dce808ab13de

Why the hell did the model feed the title as 3.5 Sonnet?!?!?! Am I using 3.5 Sonnet by mistake?