The official release table shows each model's best score across effort settings: xhigh for Claude Fable 5.2 and Fable 5.1, max for Claude Opus 5.5 and GPT-6 Astra. Fable 5.2's max-setting score is 45.7, including 22 graded timeouts.Cognition's agentic coding benchmark. Version 2.0 is a rebuilt 132-task suite (1.1 had 150) graded on whether code changes could be merged directly from the maintainer's perspective; 2.0 scores are not comparable with 1.1.