EEBench Released: AI Circuit Design Capability Evaluation, Claude Opus 5 Takes First Place with 61.6%

EEBench, a benchmark designed to objectively evaluate the electronic engineering design capabilities of AI models, has been released. This benchmark is designed to allow agents to directly handle components, connections, and electrical constraints by using atopile, which represents circuits as declarative code rather than GUI-based CAD tools.

Evaluation Method and Features

EEBench verifies design accuracy through simulation and design checks. Each requirement generates measurement items with thresholds, measuring voltage drop, effective capacitance, transient response, etc., via SPICE simulation. Notably, it uses actual manufacturer components, requiring consideration of trade-offs between component tolerance, cost, and availability. Scoring is fully deterministic, and cost efficiency is only reflected in the score if the circuit functions.

Key Model Performance Rankings (As of September 1)

The performance of major AI models on the 13 tasks of EEBench V1 is as follows:

• Claude Opus 5: 61.6% (1st)
• Grok 4.6: 57.1% (2nd)
• Claude Fable 5.1: 56.4% (3rd)
• Claude Fable 5: 54.3% (4th)
• Claude Opus 4.8 Max: 51.4% (5th)
• GPT-5.5: 42.3%
• GPT-5.6 Sol: 39.4%

xAI included EEBench in the Grok 4.6 model card, and Anthropic models consistently showed good performance in this environment. OpenAI's GPT-6 Astra has not yet released EEBench results, but demonstrated interest in the electronics field through a KiCad demo.

Implications and Limitations

Whether AI can design circuit boards is partially 'yes'. However, EEBench does not yet evaluate layout, manufacturing, or product bring-up capabilities, and challenges remain in ensuring stability in complex analog designs or worst-case tolerance corners. This benchmark is interpreted as an early sign that AI labs are beginning to take electronic engineering seriously.