OpenAI says its models are being tested on hardware design with Mazda and a racing team

OpenAI published a post on 3 November 2026 about using its models to design, simulate and build physical products. It says OpenAI's hardware and infrastructure teams use Codex and ChatGPT for computer-aided design, manufacturing drawings, simulations and design review, across robotics and compute infrastructure. It says that since February, model usage across those hardware teams, measured in tokens, has grown 23-fold.

The post describes a test in which GPT-6 was used to define requirements and GPT-6 Astra was used on the enclosure, the circuit board and the software for a device that talks to Codex. It says two team members designed that test device in under a week as a side project, and a contract manufacturer built a small batch of prototypes in under a month. It says getting the device ready for manufacturing still took several rounds of review, and that Astra did not consistently account for the manufacturer's production limits, tolerances or assembly details.

On a benchmark it calls BenchCAD 2.0, for reconstructing 3D parts and assemblies and identifying connections in circuit boards, the post says a newer internal model scored 0.875 on average, compared with 0.594 for GPT-6 Astra at its highest reasoning setting. It says OpenAI is expanding a collaboration with Chip Ganassi Racing to test workflows that turn 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, with the team's engineers reviewing the outputs. It also says it is starting a research collaboration with Mazda, with Mazda's aerodynamics and crash-safety engineers, using simulations of airflow and structural behaviour. It says more progress will be shared in the coming months. The post does not state a price or a release date for a new model.

What it means for business: this is a research and collaboration post, not a new model on the API and not a safety rating. The 23-fold figure is token use inside OpenAI's own hardware teams. The benchmark scores are OpenAI's. The Mazda and racing-team work is described as testing, with the partners' engineers reviewing the outputs. This describes the post. It is not advice on a design tool.