CommunityBench - Preference Distribution Kendall Tau: leaderboard
Rank agreement between a model's predicted split of a Reddit community's votes across candidate replies and the vote-derived split, as Kendall's tau (-1 to 1, higher is better). It uses the same predictions as the JSD board (CommunityBench, Jiayu Lin and Zhongyu Wei, Fudan University, 2026).
Metric: Kendall's Tau (-1 to 1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4 | 0.17 | #306 |
| 2 | Qwen 3 32B | 0.16 | #960 |
| 3 | DeepSeek V3 (0324) | 0.15 | #704 |
| 4 | GPT-4o | 0.14 | #741 |
| 5 | Qwen 2.5 72B Instruct | 0.14 | #1093 |
| 6 | Qwen 3 14B | 0.11 | #1235 |
| 7 | GLM-4 32B (0414) | 0.11 | #1407 |
| 8 | Llama 3.3 70B Instruct | 0.11 | #1353 |
| 9 | Llama 3.1 70B Instruct | 0.08 | #1541 |
| 10 | Qwen 2.5 14B Instruct | 0.08 | #1745 |
| 11 | Qwen 2.5 7B Instruct | 0.08 | #2583 |
| 12 | glm-4-9B (0414) | 0.06 | #2196 |
| 13 | Llama 3.1 8B Instruct | 0.04 | #4277 |
| 14 | Mistral 7B Instruct (v0.3) | 0.04 | #6698 |
| 15 | Qwen 3 8B | 0.04 | #1838 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=communitybench-preference-distribution-kendall-tau · How It Works · Data refreshed daily, snapshot 2026-11-12.