Livenerf launches benchmark to track whether Claude Opus 5.5 quietly degrades after release

Livenerf is a new open-source benchmark designed to test claims that Anthropic models get worse after launch. It starts tracking Claude Opus 5.5 from release day using frozen prompts, pinned tooling, raw logs, and statistical analysis to detect model drift over thousands of samples.

The project reflects growing demand for independent, longitudinal AI evaluations as model providers can change routing, inference settings, or serving infrastructure without renaming products. If widely adopted, post-launch benchmarking could pressure labs to offer more transparency around model updates and performance stability.