OpenAI has dissolved its standalone robotics research team, reassigning staff into the company's broader research and applied organizations, Fortune reported on November 13, 2026. The company says the move reflects its belief that general-purpose AI models can now steer physical robots directly, without a dedicated unit building specialized software underneath them.
Researchers who worked on core robotics science now report to Chief Research Officer Mark Chen as part of OpenAI's main research team. Staff who worked on applied robotics projects and hardware have moved into the company's Applied organization, which reports to President Greg Brockman. OpenAI describes the change as an internal reorganization rather than a retreat, and says its target of a public humanoid robot demonstration in 2027 has not changed.
Why fold the team instead of growing it?
The restructuring rests on a technical claim: that OpenAI's general-purpose model, GPT-6 Astra, already performs competitively against robotics-specific systems. According to an Nvidia study cited by Fortune, Astra-powered agents reached a 74.8% success rate on simulated manipulation tasks, compared with 51.0% for models built specifically for robotics.
If a model trained mostly on text, images and video can out-perform a purpose-built robotics model, the case for a separate team weakens. OpenAI's internal framing is that robotics has become a downstream application of frontier intelligence rather than a distinct research problem requiring its own roadmap, headcount and compute budget.
This is not OpenAI's first pass at robotics. The company ran an early robotics group before shutting it down in 2020, concluding that hardware and data were too limited to make progress. It rebuilt a standalone team in 2024 alongside a partnership with Figure AI, a humanoid robot maker. That partnership ended in February 2025 when Figure chose to develop its own models independently, a split that left OpenAI's internal robotics effort without a hardware partner of its own.
What does this mean for the robotics industry?
The decision lands as rivals are doing the opposite: Google DeepMind, Nvidia and humanoid makers such as Figure AI and Apptronik have all expanded dedicated robotics research in 2026, arguing that physical intelligence requires specialized training on real-world sensor and actuator data that general models lack.
Guanya Shi, an assistant professor at Carnegie Mellon University's Robotics Institute, told Fortune the move signals that OpenAI now sees robotics as "no longer a special problem" but an application layer sitting on top of its existing models. Critics could read the same move as evidence that OpenAI's robotics program struggled to produce hardware results after losing its Figure AI partnership and, more recently, after the company pulled resources from other side projects like Sora to fund its core model lineup.
Either reading points to the same open question heading into 2027: whether a chatbot-style foundation model, retrained on physical tasks, can match specialized robot brains in the real world — not just in Nvidia's simulations.