Claude Fable 5.2 Tops AA-Robotics

Fable 5.2’s result suggests frontier models are becoming materially better at translating visual reasoning into precise physical actions, though the benchmark still measures slow, expensive research-preview workflows.

• –Fable 5.2 improved dramatically over Fable 5.1’s previous 46.3 score.
• –The measure-the-box task is a meaningful vision-and-calibration challenge, not just a simple grasp or button press.
• –AA-Robotics evaluates coding agents controlling a six-axis robot arm across eight tabletop tasks.
• –Developers should view this as evidence for stronger generalist robotics agents, not production-ready autonomy.
• –The result strengthens Anthropic’s positioning of Fable 5.2 as a model for complex, long-running technical work.