OpenAI's GPT-6 Astra placed a block into a bowl in 19 of 20 robotic manipulation trials, versus 8 of 20 for Claude Fable 5.1, at roughly half the cost per run ($0.94 vs $2.12). On a puzzle-insertion task, Astra completed 2 of 20 trials, matching Fable 5.1, stalling at the same final step. Trials were graded by humans who knew the models, and the bowl comparison used different rigs.
No score is assigned. Sources and their independence are shown in the citation chain below.