Artificial Analysis: Claude Fable 5.2 leads VIALS-AA at 67%, with about 37 tool calls per task

Artificial Analysis says Claude Fable 5.2 (max with fallback) now leads VIALS-AA, its life-sciences visual-reasoning leaderboard, with 67%. It puts Gemini 4 Carbon (high) at 64% and GPT-6 Astra (max) at 63%. Gemini 4 Carbon led the board when it launched on November 6.

Artificial Analysis also reports that Fable averages about 37 tool calls per task, more than any other Anthropic model, while GPT-6 Astra averages about 8, the fewest of any model it has tested. In one protein-interaction task it describes Fable cropping the image, building a colour mask, unrolling arcs into strips, counting bars as intensity peaks and mapping its detections back onto the original to check its work. It says Fable batches calls on 28 of that task's 51 turns, averaging 1.75 calls per turn against 1.12 for Claude Sonnet 5.5.

These are Artificial Analysis's own measurements.