Can AI design circuit boards yet?
Points and comments are a snapshot, not live.
EEBench benchmark shows frontier AI models can design working circuits for defined tasks.
EEBench uses atopile to test AI models on circuit design via declarative code, not GUIs. Tasks include designing a capacitor hold-up circuit for an energy meter and a multiple-feedback low-pass filter. Checks run SPICE simulations across component tolerance corners, cost, and real manufacturer parts. Claude Opus 5 leads at 61.6%, followed by Grok 4.6 at 57.1%. xAI included EEBench in Grok 4.6's model card. OpenAI's GPT-6 Astra has not been tested yet. The benchmark aims to grade the design and verification loop, not full manufacturing.
What commenters are saying
Commenters broadly agree AI can now generate competent schematics with well-known parts and datasheets, but PCB layout remains poor. Several report successful low-complexity boards passing DRC and fabrication, while noting analog/RF designs still fail. A common tip: always feed datasheets and errata. Routing is considered easier than component placement. Two camps: those who find AI useful for board design after verification, and those who view it as still too error-prone for anything beyond hobby projects. The startup Schematik.io and the KiCAD MCP Server were mentioned as tools in this space.