Skip to content
Neural Daily

ResearchNews

OpenAI pushes Codex into CAD, PCB and simulation work

🇭🇺 Olvasd el magyarul →

OpenAI had GPT-6 Astra design a working device with no direct human edits, and an unreleased internal model scored 0.875 on BenchCAD 2.0. Mazda joins as a research partner.

By: Mikola Máté Patrik · · 2 min read

OpenAI cover: technical drawing of a round part next to the title Accelerating engineering for the physical world
Image: OpenAI · Press kit · Official social image of the OpenAI announcement · Source

What happened?

OpenAI shared (OpenAI) how its models now help engineers design, simulate and build physical products, covering mechanical and electrical design, simulation and manufacturing drawings. The company is also looking for engineering firms and software vendors to help shape training of future models.

A device with no direct human edits

As a test, OpenAI used GPT-6 Astra to build a small device for talking to Codex. The model defined the requirements and designed the enclosure, the printed circuit board, the UI and the firmware. Two team members finished it as a side project in under a week, and a contract manufacturer built prototypes in under a month.

OpenAI is candid about the gaps: getting to manufacturing still took several review rounds, because the model did not consistently account for the manufacturer's tolerances, production limits or assembly details.

Numbers and partners

  • Token usage across OpenAI's hardware teams has grown 23-fold since February, for CAD models, manufacturing drawings, simulation and FEA setup, and design reviews.
  • On the internal BenchCAD 2.0 benchmark (reconstructing 3D parts and assemblies, identifying circuit board connections), a newer unreleased internal model averaged 0.875, versus 0.594 for GPT-6 Astra at its highest reasoning setting.
  • With Chip Ganassi Racing, OpenAI is turning 3D scans of existing parts into editable CAD models and 2D manufacturing drawings.
  • A new research collaboration with Mazda covers aerodynamics and crash-safety simulation.

Why it matters

Codex has mostly been known as a software agent; OpenAI is now openly pushing it into CAD, ECAD and simulation tools. The BenchCAD jump suggests the next model will be noticeably stronger here. Hardware teams can already try Codex for drawings and simulation setup, but manufacturing tolerances still need a human check.

    Produced with the help of AI, reviewed by an editor. How we make our articles

    Reviewed and edited by the Neural Daily desk. About

    Related articles