OpenAI shares advances in using GPT-6 Astra and Codex for physical product engineering

• OpenAI is deploying GPT-6 Astra and Codex to support mechanical, electrical, and end-to-end system design, enabling engineers to handle CAD, simulations, and manufacturing drawings.
• A test device was designed by two team members with Astra in under a week, with prototypes built by a contract manufacturer in under a month, though gaps in production limits and tolerances remain.
• Model usage across OpenAI hardware teams has grown 23-fold in tokens since February, accelerating workflows for robotics and compute infrastructure.
• A newer internal model achieved an average score of 0.875 on BenchCAD 2.0, compared to 0.594 for GPT-6 Astra at its highest reasoning setting.

Industry Collaborations

• Chip Ganassi Racing is testing workflows that convert 3D scans of existing parts into editable CAD models and 2D manufacturing drawings.
• Mazda is collaborating with OpenAI to use GPT-6 Astra for aerodynamics and crash-safety simulations, exploring new nose shapes to reduce vehicle drag.

Limitations and Next Steps

• Astra did not consistently account for manufacturer production limits, tolerances, or assembly details, highlighting areas for future model improvement.
• OpenAI is seeking new collaborators in mechanical, electrical, and mechatronics fields to help shape model training and identify high-value use cases.