cobench-v2

Anthropic internally uses an unreleased model named Model 2 that scores significantly higher than Mythos 5 on the CoBench v2 benchmark, but they have no plans to release it, fueling discussions about AI advancement and singularity.