OpenAI puts ChatGPT and Codex on physical design work

OpenAI said on November 3, 2026 that ChatGPT and Codex are already in use on its hardware and infrastructure teams, and that a side project with GPT-6 Astra produced a device for talking to Codex.

The account is in OpenAI's research post, Accelerating engineering for the physical world, and in an OpenAI post on X the same day. Both are the company's own.

CAD, drawings, and simulation inside OpenAI

Hardware and infrastructure teams use Codex and ChatGPT to create and revise CAD models, prepare manufacturing drawings, set up simulations, and review designs. The work covers robotics and compute infrastructure.

Since February, model usage across OpenAI's hardware teams, measured in tokens, has grown 23-fold. OpenAI presented that as consumption. It did not pair the figure with a measured time saving.

Staff descriptions are more specific. Grace Lam, a forward deployed engineer at OpenAI, said: "Codex has dramatically accelerated how we move from an idea to an engineered part. We're iterating faster across CAD, converting 3D designs into production-ready 2D drawings, and setting up simulations and FEA with far less manual work."

Qiong Wang, a member of technical staff, said the teams use Codex and ChatGPT to organize cross-functional design constraints, explore system architectures, work across ECAD environments, and run design reviews. Wang called them dependable, around-the-clock partners, with the potential to grow into hardware specialists that can take a board from concept to Gerber output. That is stated as potential, not as current practice.

Jackie Lee, a member of infrastructure staff, said a lot of time in AEC and BIM goes into repetitive modeling, checking information across drawings and specifications, and coordinating changes between disciplines. Teams are starting to use Codex to automate routine tasks, flag inconsistencies, and turn coordination findings into issues. Architects, engineers, and BIM professionals still review and validate the results.

The Astra device, and what manufacturing caught

OpenAI tested Astra in multimodal and computer use workflows by seeing if it could create a device without any direct human edits. The device was for talking to Codex. The research post says the team used GPT-6 to define the functional requirements and design the enclosure, printed circuit board, and software integrations. That post and the X update both say Astra designed the enclosure, the circuit board, the interface, and the software.

Astra developed the look from OpenAI's recent Work Louder collaboration and from references in games and toys. It designed the enclosure and the internal structures that hold components, selected the electrical components, laid out the board to fit the enclosure, and generated the firmware and software integrations that connect the device to Codex.

Two OpenAI team members did the work in under a week, as a side project. A contract manufacturer then built a small batch of prototypes in under a month.

Production was not a clean handoff. Getting the device ready for manufacturing still required several rounds of review and iteration. Astra did not consistently account for the manufacturer's production limits, tolerances, or details required for assembly. OpenAI said those gaps are showing what future models need to do better.

BenchCAD 2.0

OpenAI also published an internal comparison on BenchCAD 2.0. The benchmark tests whether models can reconstruct 3D parts and assemblies and identify connections in circuit boards. A newer internal model scored 0.875 on average. GPT-6 Astra, at its highest reasoning setting, scored 0.594.

OpenAI called that early evidence that capability gains can affect iteration speed in professional workflows, and noted that the engineering lifecycle includes much more than design. The post does not give the number of tasks, the scoring scale, or how the internal model differs from Astra beyond those two averages.

Chip Ganassi Racing and Mazda

OpenAI said it is expanding its collaboration with Chip Ganassi Racing. With CGR's R&D team, it is testing workflows that turn 3D scans of existing parts into editable CAD models and 2D manufacturing drawings — useful when the original design files are unavailable. CGR engineers review the models and drawings for accuracy and manufacturability.

Matthew Ortman, an R&D engineer at Chip Ganassi Racing, said the model has been useful for going from raw scan data to a solid body, and that he has been working with Codex directly in CAD. Feedback he gives the model is applied, manual tweaks are still possible, and he described the result as a faster, more hands-off process for him.

The same post announces a new research collaboration with Mazda. OpenAI's research and Forward Deployed Engineering teams are working with Mazda's aerodynamics and crash-safety engineers on simulations of airflow and structural behavior, exploring performance changes while accounting for styling and cost. A caption says GPT-6 Astra helps Mazda's aerodynamics engineers explore ways to reduce vehicle drag by modeling new nose shapes, simulating airflow, and refining the most promising designs. Akihiro Nakata, a principal engineer at Mazda, said exploring and simulating dozens of variations shows which features matter for aerodynamics and where designers still have room. OpenAI said it will share more progress and results in the coming months. This post does not include measured drag or crash results.

OpenAI said a small group of industry collaborators is testing these capabilities on real engineering problems, with their engineers reviewing outputs for accuracy and practical use. It asked engineers, engineering firms, and software vendors involved in design, simulation, manufacturing, or production of complex physical products to help shape the work around the tools they use.

If you work in mechanical, electrical, or manufacturing engineering, read the November 3 post against your own release process. The claim to test is not that a model can draft an enclosure. It is whether the draft survives production limits, tolerances, and assembly details — the checks Astra missed on OpenAI's own Codex device.