OpenAI is testing its models in engineering design and manufacturing

Designing a device with Astra

According to OpenAI, its models can support the full design process for physical products. This includes preparing technical drawings, setting up simulations, and coordinating mechanical designs, electronics, software and manufacturing requirements.

The company also tested this with a system called Astra. In the experiment, GPT-6 defined the functional requirements for a device intended to communicate with Codex, then produced designs for the enclosure, printed circuit board and software integrations. Astra developed the device's appearance, designed the enclosure and internal support structures, selected the electrical components, and created the user interface, firmware and required software connections.

Two team members built the test device with Astra within a week as a side project. A contract manufacturer produced small batches of prototypes in less than a month. Preparing the design for production, however, required several rounds of checks and modifications. Astra did not always account for the manufacturer's production constraints, tolerances and assembly requirements. OpenAI says these shortcomings help identify where future models need to improve.

Model use is growing among hardware teams

OpenAI's hardware and infrastructure teams use Codex and ChatGPT to create and modify computer-aided design models, or CAD models. The models also help prepare technical manufacturing drawings, set up simulations and review designs. The work covers robotics and computing infrastructure as well.

The company says model use by its hardware teams, measured in tokens, has increased 23-fold since February. OpenAI is also examining newer models for engineering tasks. BenchCAD 2.0 tests whether systems can reconstruct three-dimensional parts and assemblies and identify connections on circuit boards.

A newer internal model achieved an average score of 0.875 on this evaluation, while GPT-6 scored 0.594 with Astra at its highest reasoning setting. OpenAI views this as an early sign that professional workflows can be accelerated, while stressing that the engineering lifecycle involves far more than design alone.

Testing with Mazda and Chip Ganassi Racing

OpenAI is examining its models on real-world engineering problems with a small group of industry partners. Participating engineers review the models' outputs to assess their accuracy and practical usefulness and to identify improvements needed for safe and reliable deployment.

The company is expanding its collaboration with Chip Ganassi Racing. With the team's research and development division, it is testing workflows that turn three-dimensional scans of existing parts into editable CAD models and two-dimensional manufacturing drawings. This could be useful when the original design files are unavailable. CGR engineers also check the models' accuracy and manufacturability.

OpenAI also announced a new research collaboration with Mazda. Together with the company's aerodynamics and crash-safety engineers, it is examining how models can assist in developing and evaluating new vehicle designs. The work is based on simulations of airflow and structural behavior while also taking performance, design and cost into account. The company says it will share further results in the coming months.

OpenAI is looking for engineers, engineering firms and software suppliers to help shape the initiative. It wants to work with partners in mechanical, electrical, electromechanical, mechatronic and related fields who can help define high-value use cases, the necessary tools and the conditions for safe industrial deployment.