OmniSch - Symbol Detection: leaderboard
Metric: Symbol detection F1 (x100): ground-truth component symbols matched by the symbol matching score of Eq. 1 (class, name, value, structure and coarse position, no box overlap required) on the industrial-EDA (Real) schematics of the OmniSch held-out test split, the model converting each PCB schematic image into a spatially weighted netlist graph with one zero-shot prompt, predicted and reference nodes aligned by maximum-weight bipartite matching; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 59 |
| 2 | Claude Sonnet 4.6 | 58.62 |
| 3 | Qwen 3.5 122B A10B | 55.84 |
| 4 | GPT-5.2 | 52.76 |
| 5 | Ministral 3 14B | 38.26 |
| 6 | GPT-5 Mini | 29.97 |
| 7 | Llama 4 Maverick | 25.27 |
| 8 | Qwen 3 VL 8B Instruct | 8.68 |
Interactive version: theaggregate.ai/benchmark?slug=omnisch-symbol-detection · How It Works · Data refreshed daily, snapshot 2026-10-07.