livenerf

About

livenerf is an open, append-only benchmark that tracks whether frontier AI models quietly lose capability after their public release, testing the recurring but previously unverified claim that providers "nerf" models post-launch. It establishes a day-0 baseline, currently for Claude Opus 5.5 (released 2026-09-22), and reruns deterministic-as-possible tests over time via headless Claude Code to catch regressions from causes like quantization, model swaps, or reduced-effort routing.