SupraBench - Host-Guest Description (CoT): leaderboard

Metric: ROUGE-1 F1 x100 (0-100) of open-ended descriptions of candidate hosts or binding guests against expert references (135 questions), chain-of-thought prompt (explicit reasoning before the answer). Source: arxiv.org. Saturation forecast: Around 2041. 8 models tracked.

Top models

#ModelScore
1Llama 3.1 8B Instruct14.4
2GPT-5.4 Mini12.9
3Gemini 3 Flash11
4GPT-5.4 Nano9.8
5Llama 3.1 70B Instruct9.7
6Qwen 3.5 27B9.2
7Qwen 3.5 9B6.8

Interactive version: theaggregate.ai/benchmark?slug=suprabench-host-guest-description-cot · How It Works · Data refreshed daily, snapshot 2026-09-29.