CommunityBench - Community-Consistent Generation: leaderboard
Metric: BTL-Elo rating from majority-voted LLM-judge pairwise comparisons. Source: arxiv.org. Saturation forecast: Estimated already saturated. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek R1 0528 | 812.3 |
| 2 | Grok 4 | 478.67 |
| 3 | Qwen 3 14B | 272.39 |
| 4 | GPT-4o | 256.28 |
| 5 | Qwen 3 32B | 237.8 |
| 6 | GLM-4 32B (0414) | 233.49 |
| 7 | DeepSeek V3 (0324) | 206.22 |
| 8 | GLM-4 9B (0414) | 176.85 |
| 9 | Qwen 3 8B | 87.77 |
| 10 | Llama 3.1 70B Instruct | -54.9 |
| 11 | Llama 3.3 70B Instruct | -56.64 |
| 12 | Qwen 2.5 72B Instruct | -149.18 |
| 13 | Qwen 2.5 14B Instruct | -154.87 |
| 14 | Llama 3.1 8B Instruct | -166.28 |
| 15 | Qwen 2.5 7B Instruct | -262.68 |
Interactive version: theaggregate.ai/benchmark?slug=communitybench-community-consistent-generation · How It Works · Data refreshed daily, snapshot 2026-09-25.