Claude Opus 5.5 scores 76.28 and GPT-6 Astra 75.99 in RoboColiseum's reported four-model comparison. Their close overall scores conceal a larger gap between understanding instructions and executing manipulation reliably. The RoboColiseum team introduces Agent2Act, a shared
RoboColiseum Compares Language Models on Instruction Following and Robot Manipulation
By
–
