OmniSch - Graph Edit Similarity: leaderboard
Metric: One minus normalized graph edit distance (x100) between the predicted and ground-truth netlist graphs 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 July 2027. 9 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Sonnet 4.6 | 15.9 |
| 2 | Qwen 3.5 122B A10B | 14.52 |
| 3 | Ministral 3 14B | 12.23 |
| 4 | Gemini 3.1 Pro (Preview) | 10.62 |
| 5 | GPT-5.2 | 9.45 |
| 6 | GPT-5 Mini | 7.3 |
| 7 | Llama 4 Maverick | 4.34 |
| 8 | Qwen 3 VL 8B Instruct | 0.65 |
Interactive version: theaggregate.ai/benchmark?slug=omnisch-graph-edit-similarity · How It Works · Data refreshed daily, snapshot 2026-10-07.