Skip to main content
Back to timeline
OpenAISource publication:

OpenAI has Astra design a device end to end: two people, one week, a prototype — but production still needs human review

Synopsis

OpenAI reported progress in how its models support mechanical and electrical engineering and end-to-end system design, and tested it end to end: GPT‑6 Astra defined the functional requirements and designed the enclosure, printed circuit board and software integrations for a device users can talk to Codex with, without any direct human edits. Two team members designed it in under a week as a side project, and a contract manufacturer built a small batch of prototypes in under a month; getting the device ready for manufacturing still required several rounds of review and iteration, and Astra did not consistently account for the manufacturer's production limits, tolerances or assembly details.

AI-generated editorial illustration: [Skip to main content](https://openai.com/index/accelerating-engineering-for-the-physical-world/#main)

Interpretation

OpenAI's models are taking on specific parts of physical product design, from preparing drawings and setting up simulations to coordinating changes across disciplines, spanning mechanical engineering, electrical engineering and end-to-end system design. That work previously required engineers to bring mechanical designs, electronics, software and manufacturing requirements together using specialized tools. OpenAI's internal hardware and infrastructure teams use Codex and ChatGPT to create and revise CAD models, prepare manufacturing drawings, set up simulations and review designs, across robotics and compute infrastructure; since February, model usage across these teams, measured in tokens, has grown 23-fold.

The end-to-end test has the model carry the design chain from requirements to a usable interactive device, rather than producing a single drawing within it. The test used GPT‑6 to define the functional requirements and design the enclosure, the printed circuit board and the software integrations needed to connect the device to Codex. Astra developed the look of the device, the internal structures holding components in place, the electrical component selection and PCB layout and connections, the user interface, and the firmware and software integrations; two team members designed it in under a week as a side project, and a contract manufacturer built a small batch of prototypes in under a month.

Measuring this capability calls for a benchmark that combines coding with engineering knowledge, not only general multimodal testing. BenchCAD 2.0 tests whether models can reconstruct 3D parts and assemblies and identify connections in circuit boards. A newer internal model averaged 0.875 on that benchmark, compared with 0.594 for GPT‑6 Astra at its highest reasoning setting.

Perspective

The demonstrated scope is OpenAI's internal hardware and infrastructure teams using Codex and ChatGPT for CAD modeling, manufacturing drawings, simulations and design review; one device prototype designed by two team members as a side project; and testing on real engineering problems with a small group of industry collaborators including Chip Ganassi Racing and Mazda, whose engineers review model outputs for accuracy and practical use, with Chip Ganassi Racing engineers also reviewing models and drawings for accuracy and manufacturability.

Getting the device ready for manufacturing still required several rounds of review and iteration, and Astra did not consistently account for the manufacturer's production limits, tolerances or assembly details; the aerodynamics and crash-safety work with Mazda is exploratory, with more progress and results to be shared in the coming months.