MolRecBench-Wild - SMILES: 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 exact SMILES matches against the expert-verified structure, with complex bond types simplified; higher is better.
Metric: SMILES exact-match accuracy (%; canonical SMILES with complex bonds simplified, samples that have a SMILES label). Source: arxiv.org. Saturation forecast: Around April 2027. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 30.06 | #262 |
| 2 | GPT-5 | 19.68 | #163 |
| 3 | Intern-S1 | 18.98 | #616 |
| 4 | Seed-1.6 (Thinking) | 15.6 | #447 (Seed-1.6) |
| 5 | GLM-4.5V | 12.13 | #739 |
| 6 | GPT-4o | 7.94 | #741 |
| 7 | Qwen-VL-Max | 6.95 | #1226 |
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-smiles · How It Works · Data refreshed daily, snapshot 2026-11-12.