OpenAI 公布 GPT-6 Astra 物理工程测试项目

OpenAI announced a set of physical-product engineering projects on November 3, 2026, including an Astra-designed test device, expanded engineering work with Chip Ganassi Racing and a new vehicle-design research collaboration with Mazda.

The capabilities span mechanical and electrical engineering and end-to-end system design, from preparing drawings and setting up simulations to coordinating changes across disciplines. The goal is to give engineers more time to explore ideas and improve designs while connecting work across specialized tools.

The device experiment tested GPT-6 Astra’s multimodal and computer-use capabilities by having the model create a device for speaking with Codex, without team members directly editing the design outputs. Astra defined the functional requirements, developed the enclosure and internal component supports, selected electrical components and designed the printed circuit board, including component placement and connections. It also designed the user interface and generated the firmware and software integrations needed to connect the device to Codex.

In an earlier Astra demonstration, the model placed components and routed copper connections in KiCad, converting an electronic schematic into a printed-circuit-board layout.

Two team members designed the test device with Astra in under a week while working on it as a side project. A contract manufacturer subsequently produced a small batch of prototypes in under a month.

Manufacturing readiness still required several rounds of review and iteration. Astra did not consistently account for the manufacturer’s production constraints, tolerances or assembly requirements. Those shortcomings are being used to identify what future models need to improve, establishing a distinction between generating a complete design and preparing that design for production.

Internal Adoption and BenchCAD 2.0

Across robotics and compute-infrastructure work, OpenAI’s hardware and infrastructure teams use Codex and ChatGPT for computer-aided design modeling and revisions, manufacturing drawings, simulation setup and design reviews. OpenAI reports that model usage across its hardware teams, measured in tokens, has grown 23-fold since February. That figure tracks model consumption rather than directly measuring engineering output or time saved.

BenchCAD 2.0 provides a separate assessment of design capability, asking models to recreate 3D parts and assemblies and determine circuit-board connections. OpenAI reports an average score of 0.875 for a newer internal model, compared with 0.594 for GPT-6 Astra at its highest reasoning setting, and presents the gain as early evidence that AI models could accelerate professional iteration while noting that design is only one part of the engineering lifecycle.

That leaves a clear gap between benchmark performance and production readiness: the evaluation covers design-related tasks, while the device experiment shows how manufacturing requirements can still necessitate further review.

OpenAI is testing these capabilities with a small group of industry collaborators whose engineers review model outputs for accuracy and practical use. Their assessments are intended to identify which tasks models can support and what must improve for safe, reliable deployment.

At Chip Ganassi Racing, the expanded collaboration tests workflows that convert 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, particularly when the original design files are unavailable. CGR engineers check the resulting models and drawings for accuracy and manufacturability, supplying feedback on where manual work can be reduced and where the models need improvement.

During its first year, the Chip Ganassi Racing partnership concentrated on interpreting data from car sensors, races and pit-crew performance. The new workflows add physical-part engineering and prototyping to that earlier data work.

The Mazda collaboration pairs OpenAI’s research and Forward Deployed Engineering teams with Mazda aerodynamics and crash-safety engineers. The teams are using airflow and structural-behavior simulations to develop and evaluate vehicle designs while accounting for styling and cost. In the aerodynamics work, GPT-6 Astra generates alternative nose shapes, runs airflow simulations and refines promising designs in an effort to reduce drag.

Akihiro Nakata, a principal engineer at Mazda, said simulating dozens of variations helps engineers understand which features matter for aerodynamics and where designers have room to explore. OpenAI plans to share further progress and results from the research.

OpenAI is also inviting engineering firms and software vendors involved in design, simulation, manufacturing and production to propose high-value uses and tooling for future collaborations. The central question across these projects is whether better design outputs can carry through production constraints and engineering review, the stages that still determine whether a generated design becomes a manufacturable product.