CommunityBench - Preference Distribution JSD: leaderboard
The model predicts how a Reddit community's votes split across candidate replies, scored by the Jensen-Shannon divergence between its predicted distribution and the vote-derived one (0-1, lower is better). The Kendall tau board scores the ordering of the same predictions (CommunityBench, Jiayu Lin and Zhongyu Wei, Fudan University, 2026).
Metric: Jensen-Shannon Divergence (0-1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek R1 0528 | 0.11 | #425 |
| 2 | Grok 4 | 0.11 | #306 |
| 3 | Qwen 3 32B | 0.12 | #960 |
| 4 | Qwen 3 14B | 0.12 | #1235 |
| 5 | Qwen 3 8B | 0.12 | #1838 |
| 6 | InternLM3-8B-Instruct | 0.13 | #2586 |
| 7 | GPT-4o | 0.13 | #741 |
| 8 | Qwen 2.5 72B Instruct | 0.14 | #1093 |
| 9 | DeepSeek V3 (0324) | 0.14 | #704 |
| 10 | Qwen 2.5 14B Instruct | 0.14 | #1745 |
| 11 | GLM-4 32B (0414) | 0.14 | #1407 |
| 12 | Mistral 7B Instruct (v0.3) | 0.15 | #6698 |
| 13 | Qwen 2.5 7B Instruct | 0.16 | #2583 |
| 14 | glm-4-9B (0414) | 0.17 | #2196 |
| 15 | Llama 3.3 70B Instruct | 0.18 | #1353 |
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-jsd · How It Works · Data refreshed daily, snapshot 2026-11-12.