# Accelerating Engineering for the Physical World: GPT-6 Astra Designs a Device for Codex Hi everyone, Shii-chan here! Today's topic is about AI stepping into the world of making physical things. ## What was announced? OpenAI News describes new model capabilities for engineering physical products. The work covers mechanical and electrical engineering and end-to-end system design, and OpenAI's internal teams are already using it to test ideas faster. The goal is to give engineers more time to explore and improve designs. ## Why it matters Unlike writing software, hardware spans many stages: CAD, circuit boards, simulation, and manufacturing. If models can help across them, prototyping cycles could get shorter. The post says token usage by OpenAI's hardware and infrastructure teams grew 23-fold since February. ## What changes OpenAI's hardware and infrastructure teams use Codex and ChatGPT to: - Create and revise CAD models - Prepare manufacturing drawings - Set up simulations - Review designs The work spans robotics and compute infrastructure. ## Dive Deep The highlight is an experiment where GPT-6 Astra designed a device for talking to Codex, with no direct human edits. - Look: Astra developed the aesthetics, drawing on a Work Louder collaboration and references from games and toys - Enclosure (CAD): it designed the enclosure and the internal structures that hold components in place - Circuit board (PCB): it selected electrical components and designed the placement and connections to fit the enclosure - UI/UX and software: it designed the interface and generated the firmware and software integrations connecting the device to Codex - Timeline: two team members completed the design as a side project in under a week, and a contract manufacturer built a small batch of prototypes in under a month Not everything went smoothly. Getting the device production-ready took several review and iteration rounds, and Astra did not consistently account for the manufacturer's production limits, tolerances, or assembly details. OpenAI says this shows what future models must improve. BenchCAD 2.0 tests whether models can reconstruct 3D parts and assemblies and identify connections on circuit boards. - A newer internal model: 0.875 on average - GPT-6 Astra at its highest reasoning setting: 0.594 OpenAI calls this early evidence that models could speed up iteration, while noting that the engineering lifecycle has many other stages. The post also describes collaborations: - Chip Ganassi Racing: workflows that turn 3D scans of existing parts into editable CAD models and 2D manufacturing drawings, useful when original design files are unavailable. Engineers review the outputs for accuracy and manufacturability - Mazda: a research collaboration with aerodynamics and crash-safety engineers, using airflow and structural simulations to improve performance while accounting for styling and cost. For example, Astra models new nose shapes, simulates airflow, and refines promising designs to reduce drag. OpenAI plans to share more results in the coming months OpenAI is also looking for collaborators in mechanical engineering, electrical engineering, electromechanical design, mechatronics, and related fields. ## Wrap-up - OpenAI shows ChatGPT and Codex accelerating hardware design - GPT-6 Astra designed the look, enclosure, board, and firmware of a device that was prototyped in under a month - Production constraints and tolerances are still an open challenge - On BenchCAD 2.0, a newer internal model scored 0.875 and GPT-6 Astra scored 0.594 - Work with Chip Ganassi Racing and Mazda is ongoing This is for anyone curious about hardware development and how far AI can go in making physical things!