OpenAI Explores Model Capabilities Across Physical Engineering, Hardware, and AEC Workflows
Frontier AI models · AI assistants · BIM collaboration
OpenAI reported advances in using frontier models such as GPT-6 and Astra to support physical product design, CAD and BIM tasks, and multidisciplinary coordination. While internal testing and industry collaborations demonstrated faster iteration in mechanical design and simulation setup, OpenAI noted that models still showed limitations in manufacturing tolerances, production limits, and assembly details.
On November 3, 2026, OpenAI reported on internal progress and industry collaborations applying frontier models to physical engineering, spanning mechanical and electrical engineering, compute hardware, robotics, and AEC/BIM workflows. In a test project, two team members used GPT-6 and Astra to design an enclosure, PCB, and firmware for a Codex device in under a week, followed by small-batch contract manufacturing. OpenAI noted that Astra did not consistently account for assembly details, tolerances, or manufacturer limits, requiring iterative review rounds. Internally, hardware and infrastructure teams used Codex and ChatGPT for CAD revision, manufacturing drawings, and simulations, reporting a 23-fold increase in token usage across hardware teams since February. An OpenAI infrastructure staff member noted using Codex in AEC and BIM to automate routine tasks, flag drawing inconsistencies, and document coordination issues for review by architects and engineers. In benchmarks combining coding with engineering knowledge, an internal model achieved an average score of 0.875 on BenchCAD 2.0 compared with 0.594 for GPT-6 Astra. OpenAI is also testing workflows with Chip Ganassi Racing to turn 3D scan data into editable CAD models, and collaborating with Mazda on aerodynamics and crash-safety simulations.
Key takeaways
- OpenAI tested Astra and GPT-6 to design a hardware device in under a week, but noted the model did not consistently meet manufacturer production limits, tolerances, or assembly details.
- Hardware teams at OpenAI reported using Codex and ChatGPT across CAD and simulations, with token usage across hardware teams growing 23-fold since February.
- In AEC and BIM workflows, OpenAI infrastructure staff reported applying Codex to automate routine tasks, flag inconsistencies across drawings, and turn coordination findings into actionable issues.
- External collaborations with Chip Ganassi Racing and Mazda are testing workflows for converting 3D scan data into editable CAD models and evaluating vehicle aerodynamics and crash safety.
Source：OpenAI News Read the original ↗
Prepared, checked and automatically published by AI; not individually reviewed by a human.