Claude Fable 5.1 is Anthropic's flagship successor in the Fable line, designed as a frontier reasoning and agentic model that handles long-context professional workloads. Pre-release evaluation by Artificial Analysis showed it topping the Intelligence Index at 66 points at max effort, ahead of Claude Opus 5 and Claude Fable 5, with the strongest results yet on Terminal-Bench v2.1 at 91.4% and SciCode at 62.0%, and a notable gain on τ³-Banking. It also posted the highest measured score on HLE at 59.1%, reinforcing its positioning as Anthropic's top-tier general intelligence release, with five adjustable effort settings spanning an 11x range in output token usage and Intelligence Index scores from 58 to 66.
Beyond raw reasoning scores, Fable 5.1 is optimized for agentic knowledge work, leading Artificial Analysis's GDPval-AA v2 evaluation at 1,853 Elo and tying at the top of AA-Briefcase at roughly 1,694 Elo, where it outscores peers on analytical quality and rubric correctness. It also attempts more AA-Omniscience questions than Claude Opus 5 and reaches the highest accuracy measured at 67.2%. Anthropic kept standard input and output pricing unchanged from Fable 5 while cutting cache read pricing by 75%, materially lowering cost on agentic workloads where cached context dominates, which makes the model a practical fit for sustained coding assistants, multi-step tool-driven workflows, and long-document analysis when deployed through GitLab Duo.
