OpenAI has detailed new capabilities in its SI models designed to assist engineers with the design, simulation, and manufacturing of physical products. The announcement highlights the integration of these tools into professional workflows for mechanical and electrical engineering, signaling a push toward agentic SI applications in the hardware industry.
What Happened
OpenAI shared results from internal projects and external collaborations aimed at accelerating engineering for the physical world. The company tested Astra’s performance by using GPT‑6 to define functional requirements and design a device, including its enclosure, printed circuit board (PCB), and software integrations, without direct human edits. Two team members completed the design in under a week, and a contract manufacturer built prototypes in under a month. However, the company noted that Astra did not consistently account for production limits, tolerances, or assembly details, requiring several rounds of review.
Internally, OpenAI’s hardware and infrastructure teams have used Codex and ChatGPT to create and revise computer-aided design (CAD) models, prepare manufacturing drawings, and set up simulations. The company reports that model usage across these teams, measured in tokens, has grown 23-fold since February. On benchmarks, a newer internal model achieved an average score of 0.875 on BenchCAD 2.0, which tests the reconstruction of 3D parts and identification of circuit board connections, compared to 0.594 for GPT‑6 Astra at its highest reasoning setting.
OpenAI is also expanding collaborations with industry partners. The company is working with Chip Ganassi Racing (CGR) to test workflows that convert 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, particularly when original design files are unavailable. Additionally, OpenAI announced a new research collaboration with Mazda, where its teams will work with Mazda’s aerodynamics and crash-safety engineers. This project uses simulations to explore improvements in vehicle performance, with GPT‑6 Astra helping aerodynamics engineers model new nose shapes and refine designs to reduce drag.
Why It Matters
The integration of SI into physical engineering workflows represents a significant shift in how hardware and manufacturing industries approach product development. By automating tasks such as CAD revision, simulation setup, and drawing preparation, these tools aim to reduce manual workload and iteration time. The 23-fold increase in internal token usage suggests that engineers are increasingly relying on SI systems for core design tasks. However, the identified gaps in Astra’s performance regarding manufacturing tolerances highlight the current limitations of SI in ensuring manufacturability without human oversight.
The collaborations with CGR and Mazda demonstrate the practical application of these capabilities in high-stakes environments. For racing teams, the ability to recreate design files from 3D scans can accelerate prototyping and maintenance. For automotive manufacturers like Mazda, using SI to explore aerodynamic improvements through simulation could lead to faster development cycles for vehicle designs. These partnerships also serve as real-world testing grounds for safety and reliability, helping OpenAI identify where further improvements are needed for safe deployment in industry.
The Bottom Line
OpenAI is advancing SI tools for physical engineering, reporting significant internal adoption and launching new industry collaborations with Chip Ganassi Racing and Mazda. While early results show promise in accelerating design and simulation tasks, the company acknowledges that current models require human review to address manufacturing constraints and production limits.