AI has spent years getting better at writing text and code. OpenAI now says it is moving into the physical world. In a post published on November 3, 2026, titled “Accelerating engineering for the physical world”, the company described how its models help engineers design, simulate and build physical products. This article summarizes OpenAI’s own claims, which have not been independently verified.
What OpenAI announced
OpenAI says its models can now support mechanical and electrical engineering and end-to-end system design. The company says it is teaching models to help with work such as preparing drawings, setting up simulations and coordinating changes between disciplines, and that its goal is to give engineers more time to explore ideas.
The device test
The most eye-catching example is a device built with GPT-6 Astra. OpenAI says it used GPT-6 to define the functional requirements and design the enclosure, the printed circuit board (PCB) and the software integrations for a device that lets people talk to Codex, its coding tool. The company says the goal was to see if Astra could create a device without any direct human edits.
According to OpenAI, Astra handled the look of the device, the enclosure and internal structure, the choice of electrical components and the PCB layout, plus the user interface and the firmware that connects the device to Codex. Two team members worked on it in under a week as a side project, and a contract manufacturer built a small batch of prototypes in under a month.
What it did not get right
OpenAI is notably candid about the limits. It says getting the device ready for manufacturing “still required several rounds of review and iteration,” and that Astra “did not consistently account for the manufacturer’s production limits, tolerances, or details required for assembly.” The company says those gaps help show what future models need to do better.
Inside OpenAI’s hardware teams
The post says OpenAI’s hardware and infrastructure teams use Codex and ChatGPT to create and revise computer-aided design (CAD) models, prepare manufacturing drawings, set up simulations and review designs. OpenAI says model usage across its hardware teams, measured in tokens, has grown 23-fold since February. The post includes quotes from OpenAI staff who describe faster iteration, and it is worth remembering that these are statements from the company’s own employees.
Benchmark claim
OpenAI also reports a result on BenchCAD 2.0, which tests whether models can reconstruct 3D parts and assemblies and identify connections in circuit boards. It says a newer internal model scored an average of 0.875, compared with 0.594 for GPT-6 Astra at its highest reasoning setting. The newer model is not publicly available, and the benchmark results come from OpenAI itself.
Industry partners
The company says it is working with a small group of industry collaborators, whose engineers review the models’ outputs for accuracy and practical use.
- Chip Ganassi Racing: OpenAI is expanding its collaboration to test workflows that turn 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, which is useful when the original design files are missing.
- Mazda: OpenAI announced a new research collaboration with Mazda’s aerodynamics and crash-safety engineers. They are using simulations of airflow and structure to explore improvements in performance while accounting for styling and cost. OpenAI says it will share more results in the coming months.
What to keep in mind
This announcement is a company blog post, not a peer-reviewed study. The device was a side project, the manufacturing readiness problems were real and the benchmark involves an internal model. At the same time, the post offers a useful picture of where AI is heading: from helping write code to helping coordinate the many steps, tools and tradeoffs of building physical things.
For engineers, the practical takeaway is that tools which handle drawings, simulations and routine coordination may soon become part of everyday workflows, with humans still reviewing the results. OpenAI itself says engineers review outputs and that some tasks still need improvement.
What happens next
OpenAI is inviting engineers, engineering firms and software vendors involved in the design, simulation, manufacturing or production of complex physical products to work with it. Watch for the promised Mazda results and for independent testing from engineering teams outside OpenAI.
The bottom line
OpenAI’s message is that AI is becoming a useful partner in building physical products, and that it is still far from a hands-off solution. The most credible part of the announcement may be the company’s own admission of what still goes wrong.
*Source: OpenAI, linked above. All claims are the company’s own, as published on November 3, 2026.*
Image: Circuit board with protective layer ready for etching by Tinux via Wikimedia Commons, CC0 1.0
