SupraBench - Top-Binder Selection (Few-Shot): leaderboard
Top-binder selection in SupraBench (2026), a supramolecular chemistry benchmark: accuracy in percent at choosing the strongest-binding of four candidate guests for a host, with three to five in-domain demonstrations. Higher is better.
Metric: Accuracy (%) at picking the strongest of four candidate guests for a host (2,264 questions), three to five in-domain demonstrations. Source: arxiv.org. Saturation forecast: Around February 2028. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 51.3 | #164 |
| 2 | GPT-5.4 Mini | 43 | #389 |
| 3 | GPT-5.4 Nano | 41.9 | #635 |
| 4 | Qwen 3.5 27B | 39.2 | #359 |
| 5 | Qwen 3.5 9B | 37 | #803 |
| 6 | Llama 3.1 70B Instruct | 35.4 | #1546 |
| 7 | Llama 3.1 8B Instruct | 28.3 | #4246 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=suprabench-top-binder-selection-few-shot · How It Works · Data refreshed daily, snapshot 2026-11-13.