OpenAI Says GPT-6 Astra Designed a Codex Hardware Device in Under a Week

OpenAI said the model carried out every stage of the design. That covered the enclosure, the printed circuit board layout, including choosing components and routing traces, as well as the firmware and the user interface. A contract manufacturer then built a small batch of prototypes in under a month.

The process was not error-free. Several rounds of revisions were needed, The Decoder reported, because Astra did not always take account of manufacturing constraints, tolerances and assembly details. Independent engineers have raised similar concerns before: earlier coverage of Astra's design abilities reported that engineers testing its output found most generated mechanical parts were not safe to build.

A Stronger Internal Model

OpenAI also pointed to a large jump in performance from an unreleased system. It said a newer "internal model" scored 0.875 on the BenchCAD 2.0 benchmark, compared with 0.594 for GPT-6 Astra at its highest reasoning setting. The benchmark tests whether models can rebuild 3D parts and assemblies and identify connections on circuit boards. OpenAI has not disclosed any details about the internal model.

Inside the company, OpenAI said its hardware teams use Codex and ChatGPT to produce CAD models, manufacturing drawings and simulations in areas including robotics and compute infrastructure, and that their token usage has risen 23 times since February.

The claims are OpenAI's own and have not been independently verified. They come as the company works on its own consumer hardware, which has been reported as something between an AI speaker and a smartphone. For manufacturers and hardware start-ups, the account suggests AI models are beginning to compress early design cycles, while showing that human review of manufacturability is still needed before production.