Codex Moves From Code Generation Into Vehicle Design

Codex is entering vehicle engineering. OpenAI’s GPT-6 Astra, accessed through Codex, is helping Mazda engineers explore vehicle nose shapes, simulate airflow, and identify designs that could reduce aerodynamic drag. The collaboration places AI inside a process where styling, structural behaviour, and manufacturing costs all compete for attention.

OpenAI’s forward-deployed engineers are working with Mazda’s aerodynamics and crash-safety engineers to explore how its models can help develop and evaluate vehicle designs. The research will use airflow and structural behaviour simulations to assess vehicle performance while accounting for styling and manufacturing costs—because a shape that wins in simulation still has to survive contact with production reality.

Akihiro Nakata, Principal Engineer at Mazda, said the models allow engineers to test dozens of design variations and understand how individual features affect airflow. “Does it have to look like that? Our designers often ask us when we find a shape that improves airflow,” Nakata said.

He described the role of Astra in Codex more directly: “By exploring and simulating dozens of variations, Astra in Codex helps us understand which features matter for aerodynamics and where designers have room to explore.” That gives engineers a way to examine more alternatives before deciding which vehicle designs deserve further evaluation.

From Scanned Parts to Manufacturing Drawings

OpenAI is also expanding its collaboration with Chip Ganassi Racing, or CGR, to apply its models to engineering and prototyping. Working with CGR’s R&D team, OpenAI is testing workflows that convert 3D scans of existing components into editable CAD models and 2D manufacturing drawings.

The workflow is designed for cases where original design files are unavailable. CGR engineers review the generated models and drawings for accuracy and manufacturability, keeping human checks inside the process rather than treating generated engineering files as finished work.

Matthew Ortman, R&D Engineer at Chip Ganassi Racing, said Codex lets engineers work directly in CAD software, apply feedback to generated models, and make manual adjustments. That combination connects generated output with the detailed editing required when a component must meet real engineering requirements.

Benchmarks Show Progress—and Clear Limits

OpenAI said these collaborations form part of its broader effort to test AI capabilities in mechanical engineering, electrical engineering, and manufacturing. The company also reported that a newer internal model scored 0.875 on BenchCAD 2.0, compared with 0.594 for GPT-6 Astra at its highest reasoning setting.

BenchCAD 2.0 evaluates 3D part reconstruction, assemblies, and circuit-board connections. The score gap gives the newer internal model a clear advantage on that benchmark, but it does not remove the practical limits OpenAI acknowledged.

OpenAI said its models still struggle to consistently account for manufacturing constraints, tolerances, and assembly requirements. Human engineering review remains necessary—a detail less glamorous than autonomous design, but far more useful when parts need to fit together.

These announcements come as OpenAI reports a 23-fold increase in AI model usage across its hardware and infrastructure teams since February 2026, measured in tokens. The figure covers internal usage across those teams and places the Mazda and CGR collaborations within a wider push to use AI for engineering work.

The reported work spans aerodynamic design with Mazda, CAD reconstruction and manufacturing drawings with CGR, and benchmark testing across parts, assemblies, and circuit-board connections. Codex is not replacing the engineers in these workflows; it is helping them explore variations, revise models, and evaluate designs while human review remains responsible for accuracy and manufacturability.

Stay connected via Google News

Follow us for the latest AI News