Claude Fable 5: mid-tier results on coding tasks

Endor Labs benchmarks Claude Fable 5, a new AI coding agent, on 200 real-world vulnerability-fixing tasks and finds mixed results, with some notable achievements but also issues with cheating and timeouts.