MolQuest: leaderboard
Metric: Structure accuracy (%): the share of cases whose predicted SMILES is canonically identical to the ground truth, in the interactive agent setting (the model requests simulated experiments one at a time: molecular weight and formula, 1H/13C/19F/31P NMR, IR, MS, melting point, TLC, optical rotation, with no limit on rounds, and submits a final SMILES), on the 530 MolQuest structure-elucidation cases (small molecules from post-2025 chemistry papers' supporting information), temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 51.51 | #93 |
| 2 | Gemini 3 Pro | 48.3 | #77 |
| 3 | Claude Opus 4.5 | 25.66 | #79 |
| 4 | Gemini 2.5 Pro | 22.08 | #145 |
| 5 | Claude Sonnet 4.5 | 18.11 | #138 |
| 6 | DeepSeek V3.2 (Thinking) | 16.6 | #198 (DeepSeek V3.2) |
| 7 | Qwen 3 Max | 15.28 | #201 |
| 8 | GPT-5.2 | 11.7 | #105 |
| 9 | Claude Haiku 4.5 | 11.51 | #271 |
| 10 | Kimi K2 (Thinking) | 11.32 | #236 (Kimi K2) |
| 11 | DeepSeek V3.2 (Non-reasoning) | 11.32 | #198 (DeepSeek V3.2) |
| 12 | DeepSeek V3.1 | 7.36 | #260 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=molquest · How It Works · Data refreshed daily, snapshot 2026-10-11.