EEBench V1: leaderboard

atopile's electrical-engineering agent benchmark: agents design analog circuits for 13 private, never-published tasks as atopile code, each in its vendor's own CLI agent, and a harness grades the designs by build checks, simulation at worst-case component-tolerance corners and bill-of-materials constraints, with no human or LLM grader. The score blends 65% technical result with 35% cost efficiency against a reference bill of materials; rows are model and reasoning-effort pairs, usually three attempts per task.

Metric: Score (%): 0.65 x technical + 0.35 x cost-efficiency vs reference BOM, mean over 13 private questions. Source: eebench.org. Saturation forecast: Around February 2027. 34 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 4 CarbonHigh83.9#5
2Claude Fable 5.2xHigh77.3#1
3Claude Opus 5.5xHigh75#9
4GPT-6.2 AstraHigh74.6#4
5Gemini 4 HydrogenHigh69.8#16
6GPT-6 Astra69.3#10
7Claude Sonnet 5.5xHigh67.2#24
8Gemini 4 ArgonHigh67.1#18
9Claude Haiku 5.5High67.1#75
10GPT-6.2 SolxHigh64.2#12
11Grok 4.764#51
12GPT-6.1 SolHigh63.6#15
13Gemini 4 NeonHigh63#3
14Muse WonderHigh62.7#34
15Claude Opus 561.6#26

Interactive version: theaggregate.ai/benchmark?slug=eebench-v1 · How It Works · Data refreshed daily, snapshot 2026-11-20.