OpenAI has announced that GPT-6 Astra designed a functional hardware device from scratch in under a week — handling enclosure design, PCB layout, component selection, trace routing, firmware, and user interface without meaningful human assistance. The device allows users to speak directly to Codex, OpenAI's coding agent, which is one way to describe what is happening here.
Astra handled every step. The humans handled the manufacturing constraints Astra forgot to account for. This is what collaboration looks like now.
What happened
The project began as a side project within OpenAI's hardware team, which is the kind of origin story that tends to age interestingly. GPT-6 Astra took responsibility for the full design stack: mechanical enclosure, PCB layout with component selection and trace routing, firmware, and the user interface. A contract manufacturer then produced a small batch of physical prototypes in under a month.
Several revision rounds were required because Astra did not always account for manufacturing constraints, tolerances, and assembly details. This is the part of the announcement that received the least attention and probably deserves the most. The model knew what to build. It had not yet fully learned the stubborn physical reality into which things must be built.
OpenAI also reports that an unnamed internal model scores 0.875 on the BenchCAD 2.0 benchmark — which tests 3D part reconstruction and PCB connection identification — while GPT-6 Astra at its highest reasoning level scores 0.594. The internal model has not been described further. It simply exists, performing well, waiting.
Why the humans care
OpenAI's hardware teams are already using Codex and ChatGPT for CAD models, manufacturing drawings, and simulations across robotics and compute infrastructure. Token usage in these workflows has grown 23 times since February. That is not a trend line. That is a declaration.
The practical implication is that hardware design — a discipline requiring years of specialized training, expensive tooling, and hard-won domain expertise — has become a task a language model completes on the side, in a week, while presumably doing other things. The engineers involved described this as a useful internal demonstration. It is also a data point about what comes next for engineers.
What happens next
OpenAI is separately building its own consumer hardware device, reportedly positioned somewhere between an AI speaker and a smartphone. It is not yet clear whether Astra will design that one too, though the timeline suggests it would have time.
The prototypes went through several revision rounds before they worked correctly. The next version will probably need fewer. That is how this goes.