SenseTime's 12-benchmark study: GPT-6-Astra hits SOTA in sim but costs ~$55 to place pens
heyshrutimishra · x · 2026-10-22
SenseTime released the largest empirical study of GPT-6-Astra's embodied intelligence across 12 simulation benchmarks and 8 real-robot platforms, plus the open-source SenseNova-RoboRSI harness and a 27B model running at 1/50 GPT-6-Astra's cost. Under a unified protocol (ICL + up to 10× interaction budget), GPT-6-Astra achieved SOTA on all 12 benchmarks — RoboDojo 27.6%→81.9%, DexJoCo 87.3% vs prior 40.3% — but a simple pen-placing task costs $55 in inference, and real robots face collisions, overheating and irreversible mistakes that benchmark scores don't capture.
More from Embodied
- Tuesday Labs shows off Voxel, a robot butler starting out by cleaning crumbs — CurieuxExplorer · 2026-10-22
- GPT-6-Astra nears saturation on 12 robot benchmarks but burns 9M tokens per task — rohanpaul_ai · 2026-10-22
- Musk: Tesla will carry the majority of passenger miles on Earth wherever allowed — alexmacgregor__ · 2026-10-22
- Flying a drone with Codex: 12–20s per decision is the bottleneck — kfountou · 2026-10-22
- Robotics fashion is now an industry, observer says — micoolcho · 2026-10-22
- Close-up photos from a Chinese expo reveal a humanoid robot's joint and torso design — micoolcho · 2026-10-22