GPT-6 Astra Designs Full Hardware Device in Under a Week

Officially launched on November 4, 2026.

OpenAI announced November 4, 2026 that two hardware engineers used GPT-6 Astra to design a fully functional prototype of a Codex companion device in under a week. The push-to-talk gadget, intended for internal researcher Justin Lin to monitor coding agents without opening a laptop, represents one of the most complete hardware design workflows handled autonomously by a large language model. Astra managed enclosure design, PCB layout with component selection and trace routing, firmware, and the user interface without direct human edits during the design phase.

End-to-End Design in Seven Days

The device, internally labeled a TamaGPT experiment, features a push-to-talk puck with a small display. According to materials reviewed by R&D World, Astra generated every layer of the design autonomously, from mechanical enclosure specifications to electrical schematic routing. The team did not manually edit the designs before sending them to manufacturing, marking a significant milestone in AI-assisted hardware prototyping. OpenAI says the entire design cycle was completed in less than a week, a timeline that typically requires weeks or months with traditional workflows.

Manufacturing introduced practical constraints. Lin noted the contract manufacturer required several revision rounds because Astra did not always account for manufacturing tolerances, assembly details, and production-specific constraints. A small batch of prototypes arrived in under a month after those adjustments.

Benchmark Performance and Internal Model Comparison

OpenAI released benchmark data alongside the announcement. GPT-6 Astra, running at its highest reasoning setting, scored 0.594 on BenchCAD 2.0, a benchmark that tests whether a model can reconstruct 3D parts from multi-view renders by generating valid CAD code. An unreleased internal OpenAI model scored 0.875 on the same benchmark, suggesting future public releases may offer significantly improved hardware design capabilities. The gap between Astra and the internal model indicates ongoing development in spatial reasoning and constraint modeling.

Token Usage Surge Across Hardware Teams

OpenAI reported monthly token usage across its hardware teams has grown 23x since February 2026. The surge reflects widespread adoption of Codex and ChatGPT in CAD workflows, manufacturing drawings, and robotics simulations. The company says its engineering teams now embed AI models directly in compute infrastructure design and production planning, with token consumption accelerating month over month. The November 4 announcement comes amid a steady stream of OpenAI product updates, with 128 coding-tool stories logged in the last 90 days.

What This Means

The TamaGPT experiment demonstrates that large language models can now handle multi-disciplinary hardware design tasks with minimal human intervention during the design phase, though manufacturing still requires human oversight. The benchmark comparison suggests OpenAI is developing specialized models for engineering workflows that significantly outperform current public releases. As token usage scales across hardware teams, AI-native design pipelines may become standard practice for rapid prototyping and low-volume manufacturing. OpenAI has separately confirmed it is building a consumer device positioned between an AI speaker and a smartphone, likely leveraging similar design workflows.