OpenAI announced it has begun joint research with Mazda on AI-assisted vehicle design, with OpenAI's research division and Forward Deployed Engineering team joining Mazda's aerodynamic design and crash-safety engineers to simulate airflow around vehicles and the behavior of body structures, aiming to establish design methods that raise performance while taking styling and cost into account. In the aerodynamics work, the team uses "GPT-6 Astra," which runs on the development environment "Codex," to generate multiple proposals with different front-end shapes, analyze the airflow for each and narrow down options expected to reduce aerodynamic drag. Akihiro Nakata, a chief engineer at Mazda, said exploring and simulating dozens of variations helps understand what features matter for aerodynamics and where designers can leave room. Progress and results are scheduled to be announced within the next few months. OpenAI also disclosed performance metrics for its design-support models the same day: on "BenchCAD 2.0," GPT-6 Astra scored only 0.594 even at maximum reasoning settings, while a newer internal model recorded 0.875. Model usage in OpenAI's hardware division has grown 23-fold in token volume since February. In motorsports, OpenAI is moving to a second phase of collaboration with American racing team Chip Ganassi Racing, validating a method that generates editable CAD models and 2D manufacturing drawings from 3D scan data of existing parts. A prototype device allowing voice interaction with Codex was unveiled; two OpenAI employees designed it in less than