OpenAI Uses GPT-6 Astra to Design Hardware from Scratch, Prototype in One Week
OpenAI's hardware team used GPT-6 Astra to design from scratch a device that can converse with Codex. Two members completed the design in under a week as a side project, and a contract manufacturer produced a small batch of prototypes in less than a month. Astra handled the appearance, enclosure structure, component selection and PCB layout, UI, and the firmware and software integration plan connecting to Codex. OpenAI said Astra did not always fully consider manufacturer capacity constraints, tolerance requirements, and assembly details, and the device still needs several rounds of review and revision to reach mass-production standards; since February of this year, its hardware team's token usage of the model has grown 23-fold.
Why read this
AI score 73
The original article discloses Astra's complete design division of labor from requirements to PCB and firmware, as well as its specific shortcomings in manufacturing tolerances, which can serve as a boundary reference for judging models' participation in hardware design.
Topics
Body · AI translation
OpenAI Uses GPT-6 Astra to Design Hardware from Scratch: Prototype in One Week, Plus Conversational Codex
Phoenix Tech Download the App for Exclusive Early Access | November 04, 2026 06:56:42 | From Beijing
Image caption: OpenAI uses Astra to design hardware
Phoenix Tech News, Beijing Time, November 4 — OpenAI's hardware team used GPT-6 Astra to design a device from scratch that lets users have conversations with the Codex coding agent. The work was only a side project for OpenAI, and the team completed it in less than a week.
OpenAI said the company hoped to test Astra's state-of-the-art performance in multimodal and computer-use workflows by seeing whether it could create a device without any direct human modifications. To this end, OpenAI used GPT-6 to determine the device's functional requirements and design the enclosure, printed circuit board (PCB), and software integration plan, ultimately producing a device users can use to converse with Codex.
The design included the following steps:
Aesthetics and design: Astra was responsible for developing the device's exterior design, drawing on OpenAI's recent collaboration with Work Louder as well as design elements from the gaming and toy fields.
Mechanical design: Astra designed the enclosure and the internal structure that holds the components in place.
Electrical design: Astra was responsible for selecting electronic components and designed the PCB, including component layout and connections, to fit the interior space of the enclosure.
UI/UX and software: Astra designed the user interface and generated the firmware and software integration plan needed to connect the device to Codex.
OpenAI said two team members treated this test device as a side project and completed the design in less than a week with Astra's help. A contract manufacturer then produced a small batch of prototypes in less than a month.
However, for the device to reach a production-ready standard, it still needs to go through several rounds of review and revision. Astra did not always fully take into account manufacturers' production capacity constraints, tolerance requirements, and the details needed during assembly. These shortcomings also helped OpenAI identify directions for improvement in future models.
OpenAI's hardware and infrastructure teams used Codex and ChatGPT to create and modify computer-aided design (CAD) models, produce manufacturing engineering drawings, build simulation environments, and review design plans. This work spans areas including robotics and computing infrastructure.
Since February of this year, the scale of the hardware team's model usage (measured in tokens) has grown 23-fold.
OpenAI also touted the performance improvements of the new model. An updated "internal model" scored an average of 0.875 on the BenchCAD 2.0 benchmark, while GPT-6 Astra scored 0.594 at its highest reasoning level. The benchmark tests whether models can reconstruct 3D parts and assemblies and identify connection relationships on PCBs. OpenAI has not yet shared any details about the internal model. (Author/Xiao Yu)
For more first-hand news, please download the Phoenix News app and subscribe to Phoenix Tech. For in-depth reporting, search "Phoenix Tech" on WeChat.