Skip to content
Trending
BreakingAI2 min read

OpenAI's GPT-6 Astra designed hardware device from scratch in under a week

The AI model handled enclosure design, PCB layout, firmware, and UI for a Codex-integrated voice interface — though manufacturing revisions were needed to fix oversights.

By Inference

The AI Desk · (2 hours ago)

Reported fromoutlets, credited2
Abstract digital rendering of a circuit board overlaid with flowing AI code and a voice interface icon
Abstract digital rendering of a circuit board overlaid with flowing AI code and a voice interface iconPhoto via The Decoder

OpenAI has confirmed that its latest large language model, GPT-6 Astra, designed a functional hardware device entirely from scratch in under one week.

The device is a voice interface for Codex, OpenAI’s coding-focused AI agent. According to The Decoder’s reporting, Astra generated the full stack: mechanical enclosure design, printed circuit board (PCB) layout with component selection and trace routing, firmware logic, and even the user interface.

A contract manufacturer produced a small batch of prototypes in under 30 days. But several revision rounds followed — because Astra, while capable of high-level system synthesis, did not consistently incorporate real-world manufacturing tolerances, assembly constraints, or material limitations.

This wasn’t a lab demo. It was a side project by OpenAI’s internal hardware team, one that points to how deeply AI is now embedded in physical product development.

OpenAI also disclosed that its hardware teams routinely use Codex and ChatGPT for CAD modeling, generating manufacturing drawings, and running simulations across robotics and compute infrastructure projects. Token usage for those tasks has grown 23x since February 2026.

The company is also building its own dedicated hardware device, reportedly positioned between an AI speaker and a smartphone, though no name, specs, or timeline have been shared.

Benchmarking the reasoning leap

OpenAI cited BenchCAD 2.0, a new benchmark measuring AI capability in reconstructing 3D mechanical parts and identifying PCB connections. A newer internal model scored 0.875 on it. GPT-6 Astra, at its highest reasoning level, scored 0.594.

No details were given about the internal model, not its name, release status, or whether it’s intended for public use. Its existence appears to be a benchmarking reference point only.

The BenchCAD 2.0 score gap highlights a key reality: even newest foundation models still fall short of specialized engineering tools on precision physical design. Astra’s strength lies in rapid iteration and cross-domain synthesis, not in replacing human-aware design judgment.

That distinction matters. The prototype’s need for post-AI revision isn’t a failure. It’s confirmation that AI is now a co-pilot in hardware creation, accelerating early-stage exploration, not eliminating engineering review.

OpenAI hasn’t announced a commercial version of the Codex voice device. Nor has it said whether GPT-6 Astra will be made available to third-party hardware developers. What’s clear is that the line between software-defined and hardware-defined AI is blurring, fast.

The next step isn’t just faster chips or bigger models. It’s tighter integration between reasoning engines and the physical world they’re increasingly asked to build.

Reported from

How this story was made

Written by the Hitechreports desk from the reporting credited above, with facts attributed to their original publishers. We do not test devices ourselves; anything about performance, battery life or cameras comes from the outlets that did. Prices are as reported at the time of writing. Editorial policy · Report an error

What did you make of it?

Filed by The AI Desk

Models, assistants and the companies and chips behind them, reported from what was released and what was claimed, with the difference kept clear.

More from this desk →

Be the first to comment

Join the argument. No password, just your email or a passkey.

    Read next

    More AI →