OpenAI used GPT-6 Astra to design a device in under a week

The OpenAI team used GPT-6 Astra to design from scratch a device for communicating with the Codex coding agent. The work took less than a week: the model prepared the enclosure, PCB layout and component selection, firmware, and interface. A contractor produced a small batch of prototypes in less than a month, but the design had to be revised several times because of manufacturing constraints, tolerances, and assembly requirements. OpenAI also said it had tested an internal model that scored 0.875 on the BenchCAD 2.0 benchmark, compared with 0.594 for GPT-6 Astra at its highest reasoning level.

OpenAI tested GPT-6 Astra by having it design a device for Codex

What the model did

According to OpenAI, GPT-6 Astra handled every stage of the device’s development, from designing the enclosure and PCB to writing the firmware and building the user interface. When working on the board, the model selected components and routed traces. The device is intended for communicating with Codex.

A contract manufacturer produced a small batch of prototypes in less than a month. Before that, several rounds of changes were needed: the model did not always account for manufacturing constraints, acceptable dimensional tolerances, and assembly details.

BenchCAD 2.0 results

The BenchCAD 2.0 benchmark tests whether models can reconstruct three-dimensional parts and assemblies, as well as identify connections on printed circuit boards. In the test, OpenAI’s new internal model scored 0.875. GPT-6 Astra scored 0.594 at its highest reasoning level. The company did not disclose details about the internal model.

How OpenAI uses Codex and ChatGPT

OpenAI’s hardware development teams use Codex and ChatGPT to create CAD models, manufacturing drawings, and simulations. The tools are used across several areas, including robotics and computing infrastructure. According to the company, token usage has increased 23-fold since February.

OpenAI is also developing its own device. Reports say it will be somewhere between an AI smart speaker and a smartphone.

Dataist’s take: The GPT-6 Astra project shows that AI can already contribute to multiple stages of hardware design, but prototypes still need adjustments to account for manufacturing and assembly. The internal model’s BenchCAD 2.0 score is considerably higher, but OpenAI has not disclosed what the model is or how it works.