← The feed Many Minded · The brief
MOBILITY · forward · impact 2/5 · 2026-11-14 · OpenAI

OpenAI shifts robotics focus as general-purpose AI shows robotics advantage

OpenAI reassigns robotics researchers to core AI teams after simulation tests reveal general-purpose AI outperforms specialized robotics models in key benchmarks.

OpenAI moved robotics researchers to its main research team under Chief Research Officer Mark Chen and to Greg Brockman’s Applied organization in November 2026. The shift follows Nvidia’s analysis showing OpenAI’s GPT-6 Astra achieved 74.8% success on simulated robotics tasks—surpassing the 51% benchmark of top specialized models. On StationeryBench, Astra completed 7 out of 100 tasks across 200 trials, while competing models like Ai2’s MolmoAct2 achieved zero. Astra also demonstrated physical kitchen navigation on a humanoid robot. This follows OpenAI’s 2020 shutdown of its robotics division due to data shortages, followed by a 2024 rebuild. CEO Sam Altman aims for a public robot demo by 2027.

The simulation-based performance suggests general-purpose AI could reduce robotics deployment costs and barriers for manufacturing and logistics. If physical-world implementation follows, specialized robotics teams might become less critical as AI handles complex tasks. However, all benchmark tests occurred in virtual environments—not physical robots—meaning real-world reliability remains unproven. OpenAI’s recent CoRL booth vacating due to short-notice reorganization highlights the volatility of this transition.

What matters next: Physical-world testing progress, timeline for 2027 demos, and whether financial pressures from losing market share to Anthropic will accelerate or delay the shift. The current evidence points to potential cost reductions in industrial automation, but simulation-only results mean the path to free, accessible robotics remains untested. The source’s caveats—especially the simulation gap—must be addressed before this moves beyond the lab.

Source: The Decoder