An artificial intelligence (AI) model can work with printed circuit board (PCB) design software to arrange electronic components and create board layouts.
Developed by OpenAI, GPT-6 Astra can interact with computer applications to perform multistep tasks. Its demonstrated capabilities include working on PCB layouts using KiCad, an electronic design automation (EDA) tool. The model can operate software through its interface rather than relying only on text-based instructions or code generation.
PCB design involves translating an electronic schematic into a physical board layout. Engineers must place components, route copper traces between pins, and ensure that the design meets electrical and manufacturing requirements. This process is used in developing embedded systems, industrial controllers, communication equipment, and consumer electronics. AI-assisted interaction with EDA software could help engineers reduce manual work during board development.
However, creating a PCB layout is only part of the design process. Engineers must still check electrical connections, component placement, signal integrity, power distribution, thermal performance, and design-rule compliance. A layout must also meet manufacturing requirements before it can be sent for fabrication. The demonstration does not establish that GPT-6 Astra can independently complete all these checks or produce a manufacturing-ready PCB.
Beyond PCB design, GPT-6 Astra can interact with software to write and test code, troubleshoot problems, and complete tasks across applications. These capabilities may also support electronics development workflows that combine hardware design with embedded software development and testing.
In the OSWorld 2.0 computer-use evaluation, GPT-6 Astra scored 72.6%, compared with 65.7% for GPT-5.6 Sol. In latency simulations, it completed tasks about 47% faster, taking roughly 40 minutes per task compared with 75 minutes for its predecessor.
For computer-aided design (CAD), the model scored 95.9% on the BenchCAD evaluation when used with tools, compared with 83.3% for GPT-5.6 Sol. The evaluation measures the reconstruction of 3D objects from rendered views using CAD code; it does not directly measure PCB design accuracy.
Click here for the original announcement.
