{
  "id": 5745520,
  "title": "Can AI design circuit boards yet?",
  "url": "https://urgent.news/2026/09/04/can-ai-design-circuit-boards-yet-5745520",
  "topic": "ai",
  "section": "AI",
  "published": "2026-09-04T19:48:29.000Z",
  "source": {
    "name": "Hacker News Best",
    "slug": "hacker-news-best",
    "url": "https://eebench.org/blog/can-ai-design-circuit-boards-yet/"
  },
  "original_language": "en",
  "account": "OpenAI's GPT-6 Astra demonstrated circuit board creation in a recent launch, but measuring the effectiveness of AI-produced electronics remains a challenge. Current models have extensive knowledge of electronics, but their performance in conventional design tools is limited by the time spent navigating graphical interfaces and managing on-screen elements. EEBench, a benchmark developed by the source, tackles this issue by using declarative code to work directly with circuit components, connections, and electrical constraints.\n\nOne of EEBench's public tasks involves a residential energy meter. When the 5V supply fails, the circuit must keep the processor operational for 20ms to save accumulated readings. The protected rail must maintain voltage above the processor's 3.0V brownout threshold during this period. A real capacitor may not deliver its advertised capacitance under voltage, with tolerances, costs, and package dimensions to consider. EEBench simulates input power loss, measures voltage, effective capacitance, recovery, and package limits to evaluate the design's resilience under various conditions.\n\nHowever, electronics engineering extends beyond technical specifications. EEBench uses real manufacturer parts with datasheet specifications incorporated into SPICE models. The agent must find a combination of components that works within tolerance corners while remaining cost-effective and realistically priced. This reflects the real-world trade-offs engineers face when designing circuit boards.\n\nEEBench V1 evaluates both analog and digital designs through simulation, focusing on the requirements, design, and verification loop. While it does not yet assess layout, manufacturing, and product bring-up, the benchmark's results are promising. In the September 1 results, Claude Opus 5 scored 61.6% across 13 tasks, while Grok 4.6 ranked second at 57.1% and was closely followed by Claude Fable 5.1 at 56.4%. The inclusion of EEBench in Grok 4.6's model card by xAI highlights the growing interest in measuring a model's engineering capabilities.\n\nOpenAI's earlier models, such as GPT-5.5 with 42.3% and GPT-5.6 Sol with 39.4%, lag behind the current frontier labs. The source's enthusiasm for xAI's Grok 4.6 and the progress made by OpenAI's GPT-6 Astra on a KiCad board suggests that AI's role in circuit design is becoming an increasingly important area of research and development.",
  "summary": "Article URL: https://eebench.org/blog/can-ai-design-circuit-boards-yet/ Comments URL: https://news.ycombinator.com/item?id=49569366 Points: 243 # Comments: 147",
  "key_points": [],
  "editors_take": null,
  "illustration": null,
  "coverage": {
    "outlets": 2,
    "also_reported_by": [
      {
        "outlet": "Hacker News",
        "title": "Can AI design circuit boards yet?",
        "url": "https://urgent.news/2026/09/04/can-ai-design-circuit-boards-yet",
        "published": "2026-09-04T19:48:29.000Z"
      }
    ]
  },
  "ai_generated": true,
  "disclaimer": "Summaries, key points and the editor’s take are written by software from other outlets’ reporting and may contain errors — always check the linked original."
}