Summary
OpenAI says its models are helping internal teams with CAD, circuit-board design and simulations, and it is testing engineering workflows with industry collaborators. A device prototype and BenchCAD 2.0 results offer early examples, while manufacturing work still requires engineering review.
OpenAI says its models are helping internal hardware and infrastructure teams with computer-aided design (CAD), manufacturing drawings, simulations and design reviews. The company is also testing engineering workflows with industry collaborators, including Chip Ganassi Racing and Mazda. The examples show AI being applied across mechanical and electrical design, but OpenAI describes the work as early and says engineers review outputs for accuracy and practical use.
From device design to engineering workflows
One test involved designing a device for users to talk to Codex. OpenAI says it used GPT-6 to define the functional requirements and design the enclosure, printed circuit board (PCB) and software integrations. In its account of the work, Astra contributed to the device’s appearance, mechanical structure, electrical design, user interface, firmware and software connections.
Two team members worked on the device as a side project and completed the design in under a week. A contract manufacturer built a small batch of prototypes in under a month. Preparing the device for manufacturing took several rounds of review and iteration: OpenAI says Astra did not consistently account for production limits, tolerances or assembly details. Those are practical requirements that can determine whether a design is straightforward to build, not just whether its parts fit together in a model.
OpenAI also reports that token usage across its hardware teams has grown 23-fold since February. The company says its teams use Codex and ChatGPT for work such as revising CAD models, preparing drawings and setting up simulations. The token figure measures model usage, rather than engineering output or productivity directly.
A benchmark for selected engineering tasks
OpenAI reports that a newer internal model scored an average of 0.875 on BenchCAD 2.0, compared with 0.594 for GPT-6 Astra at its highest reasoning setting. The benchmark tests tasks including reconstructing 3D parts and assemblies and identifying connections in circuit boards.
These scores concern a defined set of design-related tasks. OpenAI notes that engineering involves more than design, and presents the results as early evidence relevant to professional workflows—not as a measure of the full engineering lifecycle.
Trials with racing and automotive engineers
With Chip Ganassi Racing, OpenAI is testing workflows that turn 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, including cases where original design files are unavailable. The racing team’s engineers review the resulting models and drawings for accuracy and manufacturability.
A separate research collaboration with Mazda is exploring the use of models in vehicle design. OpenAI says the work includes using simulations of airflow and structural behaviour to examine design options while considering performance, styling and cost. One example involves exploring nose shapes to reduce vehicle drag. Mazda engineers assess the outputs as part of the collaboration.
The trials illustrate where AI may help: converting information between engineering formats, exploring design alternatives and coordinating work across disciplines. The prototype’s manufacturing iterations and the partners’ review processes also show why engineers remain involved as these workflows are tested.