A device designed without a human edit
OpenAI said on Tuesday that two of its staff used GPT‑6 Astra to design a small device end to end as a side project, with no direct human edits. The model defined the functional requirements and produced the enclosure, the printed circuit board with its component selection and layout, the user interface, and the firmware and software integrations needed to let the device talk to Codex. The look drew on OpenAI’s Work Louder collaboration and on references from games and toys.
The two engineers finished the design in under a week. A contract manufacturer built a small batch of prototypes in under a month.
Where it broke
The company does not leave it there. “Getting the device ready for manufacturing still required several rounds of review and iteration,” the post says. “Astra did not consistently account for the manufacturer’s production limits, tolerances, or details required for assembly.”
That is the gap between a design that renders and a design a factory can build, and OpenAI names it as the thing future models need to fix.
The benchmark is OpenAI’s own
OpenAI also reports progress on BenchCAD 2.0, an internal test of whether a model can reconstruct 3D parts and assemblies and identify connections in circuit boards. A newer internal model scored an average of 0.875, against 0.594 for GPT‑6 Astra at its highest reasoning setting.
Both figures are OpenAI’s, measured on OpenAI’s benchmark, and the newer model is unnamed and unreleased. There is no independent CAD benchmark published alongside them.
Two named partners
Inside the company, token usage across OpenAI’s hardware and infrastructure teams has grown 23-fold since February, covering CAD models, manufacturing drawings, simulation setup and design review across robotics and compute infrastructure.
Outside it, OpenAI is expanding work with Chip Ganassi Racing on turning 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, useful when the original design files are gone. “I have been working with Codex directly in CAD,” said Matthew Ortman, an R&D engineer at the team.
The new name is Mazda. OpenAI’s research and forward deployed engineering teams are working with the carmaker’s aerodynamics and crash-safety engineers on developing and evaluating vehicle designs, using airflow and structural simulation. Akihiro Nakata, a principal engineer at Mazda, described using Astra in Codex to model new nose shapes and simulate airflow across dozens of variations. OpenAI says it will share results in the coming months.
What to watch
Whether the unnamed internal model ships, and whether anyone outside OpenAI measures it on a CAD benchmark OpenAI did not write. The manufacturability gap is the more honest signal: it is the part the company says its own test device failed.
