OpenAI has expanded the application of its GPT-6 Astra model to external engineering partners to automate physical design workflows. The company said Mazda’s aerodynamics engineers use the AI to simulate nose shapes to reduce drag, while the American racing team Chip Ganassi Racing uses the tool to convert 3D scans of legacy parts into manufacturing drawings and editable CAD models. These industry deployments come as the use of AI engineering tools by OpenAI’s own internal hardware teams has grown 23 fold since February.
The company recently demonstrated the model’s capabilities through a project where two engineers used Astra to design a gadget for interacting with the Codex coding agent. Astra developed the electrical components, the printed circuit board layout, and the firmware for the device in less than a week, though OpenAI noted the AI still lacks a full understanding of factory tolerances and assembly details. These results follow the announcement of a newer unreleased internal model that achieved a 0.875 score on the BenchCAD 2.0 hardware design benchmark, surpassing GPT-6 Astra’s peak score of 0.594.
