Brief IA
›

OpenAI's GPT-6 Astra Completes Full Setup in Six Days

Models & LLM·Tom Levy·

OpenAI's GPT-6 Astra Completes Full Setup in Six Days

OpenAI's GPT-6 Astra Completes Full Setup in Six Days
Key Takeaways
1GPT-6 Astra enabled OpenAI to design a device for Codex in less than a week
2A new internal model achieved 0.875 on BenchCAD 2.0, compared to 0.594 for Astra
3Token usage by hardware teams has increased 23 times since February
4A separate hardware project, between AI speaker and smartphone, is in development
Why it matters : These advancements illustrate the growing integration of AI in hardware design and the evolution of OpenAI's internal tools.

OpenAI claims to have designed, in less than a week with the help of GPT-6 Astra, a device that allows interaction with Codex, covering all stages of hardware and software design. The company also publishes scores on a CAD benchmark, reports a significant increase in internal usage of its tools, and mentions a distinct hardware project in development.

Missing Details, Significant Internal Usage Increase, and a Product in Development

OpenAI has not provided additional information about the new internal model mentioned. Furthermore, the company states that token usage by its hardware teams has increased 23-fold since February. OpenAI is also working on a hardware device in development, described as positioned between an AI speaker and a smartphone.

A Device for Codex Designed by Astra and Passed to Prototypes

The hardware team at OpenAI utilized GPT-6 Astra to develop a device intended to facilitate interaction with the coding agent Codex, a side project completed in less than a week. According to OpenAI, Astra was responsible for creating the casing, arranging the printed circuit board by selecting components and routing, as well as developing the firmware and user interface. A subcontractor then manufactured a small series of prototypes in less than a month. Several cycles of modifications were necessary, as Astra did not consistently incorporate manufacturing constraints, tolerances, and assembly requirements.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Performance Benchmarks on BenchCAD 2.0 and Tool Usage

OpenAI highlights a score of 0.875 achieved by a new internal model on the BenchCAD 2.0 benchmark, while GPT-6 Astra scores 0.594 at its most advanced reasoning level. This test evaluates the models' ability to reconstruct 3D parts and assemblies, as well as to identify connections on printed circuit boards. OpenAI indicates that its hardware teams are using Codex and ChatGPT to generate CAD models, manufacturing plans, and simulations in sectors such as robotics and IT infrastructure.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.