Opus 5.5 Under Constant Watch

🔥 What happened A new open-source benchmark called livenerf launched to track whether Claude Opus 5.5 quietly degrades after release. It runs daily for 30 days with frozen prompts and a pinned CLI to statistically measure drift. 💡 Why it matters First clean day-0 baseline against Anthropic's notorious "nerfing" rumors. The tool detects a 7.5-point accuracy drop per 10-day window for just 3.6% of the weekly plan. Output tokens reveal effort reductions far earlier than accuracy does. ⚡ Our take Finally, someone replacing vibes-based arguments with hard data. If Anthropic really degrades models post-launch, this repo will prove it—or bury the conspiracy theory for good.