> ## Content Index > Fetch the complete content index at: https://www.tao.media/llms.txt > Use this file to discover other available public pages before exploring further. # OpenAI Uses GPT-6 Astra to Design a Codex Device and Begins Mazda Engineering Research - URL: https://www.tao.media/openai-uses-gpt-6-astra-to-design-a-codex-device-and-begins-mazda-engineering-research/ - Published: 2026-11-04T21:45:29.000Z - Updated: 2026-11-04T21:47:29.000Z - Description: The projects test AI-assisted hardware design, manufacturing preparation, and vehicle simulation. - Author: Bart Hillerich - Tags: OpenAI, Mazda, AI, News OpenAI used GPT-6 Astra to design a [prototype device for talking to Codex](https://x.com/OpenAI/status/2117692475775258785?ref=tao.media), spanning its enclosure, electronics, and software, and has begun a research collaboration with Mazda on vehicle design and simulation. Two OpenAI employees developed the device with Astra in under a week as a side project, and a contract manufacturer built a small batch of prototypes in under a month. The experiment produced the designs without direct human edits, although preparing them for manufacturing required several rounds of review and iteration. The projects are part of a broader effort to apply ChatGPT and Codex to [physical engineering](https://openai.com/index/accelerating-engineering-for-the-physical-world/?ref=tao.media), including mechanical design, circuit-board development, and simulation. OpenAI is also expanding its work with Chip Ganassi Racing to turn scans of existing parts into editable engineering files. ## From Enclosure to Firmware Astra developed the device's appearance using references from games, toys, and OpenAI's recent [Work Louder collaboration](https://worklouder.cc/codex-micro?ref=tao.media). It then designed the internal structures that hold components in place, selected the electrical components, and arranged their placement and connections on the circuit board to fit inside the enclosure. 0:00 /0:19 1× [GPT-6 Astra-designed Codex device prototype | OpenAI (X)](https://x.com/OpenAI/status/2117692477947641954?ref=tao.media) The software work covered the user interface, firmware, and Codex integrations. Firmware is the code that runs on the device itself, linking its electronics to the functions users interact with. Manufacturing exposed gaps in the design process. Astra did not consistently account for the manufacturer's production limits, tolerances, or assembly requirements. Those details determine whether components can be made to the required dimensions and assembled into a working product. OpenAI is using the problems identified during review to guide improvements in future models. ## Engineering Workflows and BenchCAD Results OpenAI's hardware and infrastructure teams already use Codex and ChatGPT to create and revise computer-aided design (CAD) models, prepare manufacturing drawings, set up simulations, and review designs. The work spans robotics and compute infrastructure. Model usage across its hardware teams, measured in tokens, has increased 23-fold since February. The internal examples include individual parts and coordination across larger engineering projects. Forward deployed engineer Grace Lam reported faster CAD iteration, conversion of 3D designs into production-ready 2D drawings, and less manual work setting up simulations and finite element analysis. That analysis divides a structure into smaller elements to estimate how it responds to physical conditions such as loads. Other teams use the models to organize constraints across disciplines, explore system architectures, and work in electronic CAD environments. In architecture, engineering, construction, and building information modeling, OpenAI is beginning to automate repetitive modeling, identify inconsistencies across drawings and specifications, and turn coordination findings into actionable issues. Architects, engineers, and building information modeling professionals review and validate the results. OpenAI also reported gains on BenchCAD 2.0, which tests whether models can reconstruct 3D parts and assemblies and identify circuit-board connections. A newer internal model achieved an average score of 0.875, compared with 0.594 for GPT-6 Astra at its highest reasoning setting. The benchmark measures specific design and engineering tasks, and OpenAI treats the improvement as early evidence that models can help speed professional iteration across a broader engineering lifecycle. ## Mazda Simulations and Racing-Part Reconstruction Mazda's aerodynamics and crash-safety engineers are working with OpenAI's research and Forward Deployed Engineering teams to develop and evaluate vehicle designs. The collaboration uses airflow and structural simulations to explore performance improvements while accounting for styling and cost. In the aerodynamics example, GPT-6 Astra models new vehicle nose shapes, simulates airflow, and refines promising designs to explore reductions in drag, the aerodynamic resistance a moving vehicle encounters. 0:00 /0:37 1× [Front car aerodynamics | OpenAI (X)](https://x.com/OpenAI/status/2117692480590373245?ref=tao.media) Mazda principal engineer Akihiro Nakata said exploring and simulating dozens of variations helps the team identify which features matter for aerodynamics and where designers retain flexibility. The aim is to bring aerodynamic performance and Mazda's design requirements into the same iterative process. At [Chip Ganassi Racing](https://openai.com/index/chip-ganassi-racing/?ref=tao.media), the expanded collaboration focuses on reconstructing existing parts. OpenAI and the racing team's R&D engineers are testing workflows that convert 3D scans into editable CAD models and 2D manufacturing drawings, particularly when original design files are unavailable. That process takes scanned geometry toward a model engineers can modify and use to prepare manufacturing documentation. CGR engineers check the outputs for accuracy and manufacturability. R&D engineer Matthew Ortman said Codex has helped him move from raw scan data to a solid body while applying feedback directly within CAD. OpenAI plans to share further Mazda progress and results in the coming months. It is also [seeking engineering collaborators](https://openai.com/form/computer-use-research-collaboration/?ref=tao.media) in mechanical engineering, electrical engineering, electromechanical design, mechatronics, and related fields to identify useful applications and software tools for future model development.