Testing GPT-5.6 and Gemini Robotics Models: Impressive but Need Real Deployment Data

m_wulfmeier · x · 2026-08-04

A robotics expert evaluated Gemini-Robotics-ER-2 and GPT-5.6-Sol models while extending benchmarks.

The author praised these models for delivering the high level of accuracy and robustness required by robotics applications. However, he emphasized a core bottleneck: just as coding proficiency requires coding data, models cannot truly improve at deployment without real-world deployment data.

Related event: GPT-5.6 and Gemini Robotics Models Show Promise but Lack Real-World Generalization(3 posts)→

Original post →

More from Embodied

Embodied channel →