SpecX - Functional Group Inference (Simulated Spectra): leaderboard
Naming the functional groups present in a molecule from its spectra, zero-shot and without the molecular formula (SpecX, Xiang et al., 2026): the model reads text-encoded simulated 1H-NMR, 13C-NMR, HSQC, IR and mass spectra of molecules from the benchmark's aligned Small subset. Macro-averaged F1 from 0 to 1 over 37 SMARTS-defined functional groups on the random-split test set; higher is better.
Metric: Macro-F1 (0-1; 37 SMARTS functional-group labels, zero-shot from text-encoded simulated 1H-NMR, 13C-NMR, HSQC, IR and MS spectra of the Small subset, random split). Source: arxiv.org. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 Mini | 0.41 | #775 |
| 2 | DeepSeek V3 | 0.38 | #705 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=specx-functional-group-inference-simulated-spectra · How It Works · Data refreshed daily, snapshot 2026-11-12.