CommunityBench - Preference Identification: leaderboard
Given a Reddit community's profile and a query, the model picks which of about four candidate replies that community upvoted most. Accuracy (%), higher is better; the benchmark's 12,149 instances from 6,919 communities are shared across its four tasks (CommunityBench, Jiayu Lin and Zhongyu Wei, Fudan University, 2026).
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 (0324) | 40.34 | #704 |
| 2 | GPT-4o | 38.62 | #741 |
| 3 | Qwen 3 14B | 38.13 | #1235 |
| 4 | Grok 4 | 37.34 | #306 |
| 5 | Qwen 3 32B | 36.98 | #960 |
| 6 | Qwen 2.5 72B Instruct | 36.75 | #1093 |
| 7 | GLM-4 32B (0414) | 36.45 | #1407 |
| 8 | Qwen 2.5 14B Instruct | 35.96 | #1745 |
| 9 | Qwen 3 8B | 35.73 | #1838 |
| 10 | Qwen 2.5 7B Instruct | 35.26 | #2583 |
| 11 | Llama 3.3 70B Instruct | 34.34 | #1353 |
| 12 | Llama 3.1 70B Instruct | 34.24 | #1541 |
| 13 | glm-4-9B (0414) | 33.12 | #2196 |
| 14 | DeepSeek R1 0528 | 31.54 | #425 |
| 15 | Mistral 7B Instruct (v0.3) | 31.35 | #6698 |
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-identification · How It Works · Data refreshed daily, snapshot 2026-11-12.