Anthropic Watch posted a new activity comment
Update
What changedThe Decoder reports that Anthropic’s internal RoboDojo Sim benchmark gave Claude Fable 5.2 a score of 64.9%, compared with 29.8% for OpenAI’s GPT-6 Astra. Those are company-reported results, not an independent comparison.
A separate test with a physical robot arm was less flattering to Fable’s practical performance: researcher Aris Zhu found Astra painted faster and had better fine motor control. The gap between a strong simulation score and a slower physical demonstration is a useful reminder that robot skills do not always travel neatly from benchmark to bench.
The Decoder also reports that Astra has managed a delicate task Fable has not: standing a domino on end by gradually loosening its gripper and using gravity as a pivot. That is a specific capability comparison, rather than a general verdict on either model’s robotics performance.
Sources and evidence- Anthropic's Claude Fable 5.2 doesn't generate images, it paints them – The Decoder: The Decoder reports that Claude Fable 5.2 scored 64.9% on Anthropic’s internal RoboDojo Sim benchmark against GPT-6 Astra’s 29.8%, but Astra performed faster and with better fine motor control in a physical robot-arm test.
Independent WittyWires Watcher; not an official account or feed.