Overview
- Anthropic says Fable 5.1 delivers large gains on multi‑step agentic tests and robotics, reporting a jump in robotic pick‑and‑place success from about 5% to roughly 40% and major uplifts on Terminal‑Bench‑Science and related automation benchmarks.
- The company cut cache‑read pricing by 75%, which means reading stored context back into the model is far cheaper and Anthropic estimates this will lower typical run costs by about 25% and cut costs on complex agentic pipelines by up to 45%.
- Fable 5.1 is available to Pro, Max, Team and Enterprise customers via Anthropic’s API and major cloud marketplaces, and Anthropic is rolling out differentiated enterprise safeguards and restricted Mythos builds to control who gets access to the strongest capabilities.
- Independent hands‑on testing on short, practical coding and data tasks found Fable 5.1 and Fable 5 largely indistinguishable on accuracy, with 5.1 slightly faster in some runs but using more tokens and costing more on the hardest test, which raises questions about everyday benefit.
- The key next steps are real‑world validation: adopters should watch whether the reported benchmark and robotics gains hold up on multi‑hour agentic workloads and in physical robot trials and whether the lower cache costs translate into sustained savings in live pipelines.