MolRecBench-Wild - Graph: leaderboard
Optical chemical structure recognition on MolRecBench-Wild (Yang et al., CVPR 2026): molecular diagrams cropped from 820 recent chemistry papers, many with visual clutter or non-standard chemistry, must be transcribed from the image. Accuracy in percent of predicted CARBON graphs that match exactly in atom types, every bond type and bracketed groups, the strictest of the three tracks; higher is better.
Metric: Graph exact-match accuracy (%; CARBON graph with atom types, all bond types and bracket information). Source: arxiv.org. Saturation forecast: Around June 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 13.04 | #262 |
| 2 | GPT-5 | 8.19 | #163 |
| 3 | Seed-1.6 (Thinking) | 4.61 | #447 (Seed-1.6) |
| 4 | GLM-4.5V | 4.26 | #739 |
| 5 | Qwen-VL-Max | 3.66 | #1226 |
| 6 | Intern-S1 | 3.46 | #616 |
| 7 | GPT-4o | 2.94 | #741 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=molrecbench-wild-graph · How It Works · Data refreshed daily, snapshot 2026-11-12.