RoboColiseum robot benchmark: Claude Opus 5.5 scores 76.28, edging out GPT-6 Astra
jiqizhixin · x · 2026-11-01
The RoboColiseum team introduces Agent2Act, a shared harness connecting frontier multimodal models to real robot observations, action execution, and feedback in a closed loop.
- Models inspect images and robot state, query sensor evidence, and propose actions; the harness validates proposals, converts them into robot commands, and returns execution feedback
- Native scoring covers instruction following, spatial reasoning, manipulation, and robustness
- In a four-model comparison, Claude Opus 5.5 scored 76.28 and GPT-6 Astra 75.99 — close overall scores conceal a larger gap between understanding instructions and executing manipulation reliably
- Component comparisons show feedback and history can materially affect success rates and planning cost
More from Embodied
- Robot Intelligence Index debuts: GPT-6 Astra tops robot model ranking at 1905 — haider1 · 2026-11-01
- ETH Zurich releases EgoHTR: 150k-frame egocentric 4D dataset for humanoid terrain traversal — Scobleizer · 2026-11-01
- Dyna's long-horizon robot demo hailed as one of the cleanest yet — Scobleizer · 2026-11-01
- Tesla Robotaxi out-engages Waymo 9x on X despite smaller fleet — mitchdeg · 2026-11-01
- The Trinity: capability-driven RSI lifts frozen-model benchmarks 34 points, robot success 0→60% — YangYou1991 · 2026-11-01
- 10 spicy takes from 50 frontier robotics insiders: Astra is robotics' GPT-3 moment — ZhitingHu · 2026-11-01