OpenAI says GPT-6 Astra designed a Codex hardware prototype in under a week

OpenAI published an account on November 3 of how its models are being used for mechanical and electrical engineering, and its headline example is a small device that two employees designed with GPT-6 Astra in less than a week. The claims come from OpenAI itself and have not been independently checked.

What OpenAI says the model did

The device is a dedicated physical interface for speaking with Codex, OpenAI's coding agent, built as a side project by members of the hardware team. According to OpenAI, as reported by The Decoder and Unite.AI, Astra worked through each stage of the design: defining requirements, shaping the enclosure and internal supports, choosing electrical components, laying out the printed circuit board, writing the firmware, designing the user interface and connecting the device to Codex. A contract manufacturer then produced a small batch of prototypes in under a month. The company has not named the device or said whether it will be sold.

Where it fell short

OpenAI acknowledges that a complete design is not the same as a manufacturable one. Astra did not reliably account for the manufacturer's production constraints, part tolerances or assembly requirements, and engineers needed several rounds of review and iteration before the prototypes could be built. OpenAI says these gaps are informing work on future models. The Decoder also notes that the timeframes are OpenAI's own figures and do not show how much human review was involved.

Wider engineering push

OpenAI says its hardware and infrastructure teams now use Codex and ChatGPT to produce CAD models, manufacturing drawings and simulation setups, and that token usage across those teams grew 23-fold between February and October. On BenchCAD 2.0, an internal benchmark for reconstructing 3D parts and identifying circuit board connections, the company reports a score of 0.594 for GPT-6 Astra at its highest reasoning setting and 0.875 for a newer, undisclosed internal model.

The post also describes outside collaborations reported by Interesting Engineering: Chip Ganassi Racing is exploring how to turn 3D scans of parts into editable CAD models and drawings, and Mazda is studying aerodynamics and crash safety, with Astra exploring vehicle nose shapes and simulating airflow before Mazda engineers weigh styling and cost. These results concern design tasks; the device example shows that production readiness still depends on human engineers.