RoboColosseum benchmark: GPT-6 Astra stands out across 11,398 real-world evals

Jiafei Duan✈️CoRL2026 @DJiafeiOriginally posted 2026.11.05

Open original on X

41 days. 124 diverse sites. 11,398 real-world evaluations. At @RoboColosseum, we compared 5 VLA policies and 3 frontier AI agents in the wild. One clear takeaway: @OpenAI’s GPT-6 Astra is seriously strong. Those demos on X line up with what we’re seeing at scale. Read our 34 pages technical report for many more interesting insights: https://arxiv.org/pdf/2611.02497