CommunityBench - Preference Distribution JSD: leaderboard
Metric: Jensen-Shannon Divergence (0-1). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 0528 | 0.11 |
| 2 | Grok 4 | 0.11 |
| 3 | Qwen 3 32B | 0.12 |
| 4 | Qwen 3 14B | 0.12 |
| 5 | Qwen 3 8B | 0.12 |
| 6 | InternLM3-8B-Instruct | 0.13 |
| 7 | GPT-4o | 0.13 |
| 8 | Qwen 2.5 72B Instruct | 0.14 |
| 9 | DeepSeek V3 (0324) | 0.14 |
| 10 | Qwen 2.5 14B Instruct | 0.14 |
| 11 | GLM-4 32B (0414) | 0.14 |
| 12 | Mistral 7B Instruct (v0.3) | 0.15 |
| 13 | Qwen 2.5 7B Instruct | 0.16 |
| 14 | GLM-4 9B (0414) | 0.17 |
| 15 | Llama 3.3 70B Instruct | 0.18 |
Interactive version: theaggregate.ai/benchmark?slug=communitybench-preference-distribution-jsd · How It Works · Data refreshed daily, snapshot 2026-09-25.