New Benchmark to see if Opus 5.5 gets nerfed

A new benchmark is being conducted to determine if Anthropic's Opus 5.5 model is being "nerfed" over time. The creator will run daily deterministic tests for a month, logging discrepancies and announcing results on a public repository. The tests will measure statistical degradation in answer accuracy on a set task, acknowledging that LLM output itself is not deterministic, but the questions and graders are, aligning with standard benchmark practices.