Claude Fable 5.2 Leads VIALS-AA Life-Science Test With a 67% Score

Claude Fable 5.2 scored 67% on the VIALS-AA life-science image benchmark, while access to tools nearly doubled average performance across the systems tested. The results show that system configuration mattered alongside model selection.

The benchmark covers 161 tasks across eight domains and uses 174 images from biotech workflows, including gel blots and molecular structures.

Fable scored 67%, Gemini 4 Carbon scored 64% and GPT-6 Astra scored 63%.

Average scores nearly doubled when models had access to tools.

Tool use also varied by model. Fable averaged about 37 tool calls per task, mostly for image analysis, while Haiku averaged about 21 and GPT-6 Astra about eight. Per-task costs across the evaluated systems varied by as much as 99 times.

Anthropic announced Fable 5.2 on Nov. 9. A Nov. 10 check also placed it first on the AA-3D leaderboard at 1,271 Elo, an interactive 3D scene-generation benchmark.