CommunityBench - Preference Identification: leaderboard
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around 2033. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3 (0324) | 40.34 |
| 2 | GPT-4o | 38.62 |
| 3 | Qwen 3 14B | 38.13 |
| 4 | Grok 4 | 37.34 |
| 5 | Qwen 3 32B | 36.98 |
| 6 | Qwen 2.5 72B Instruct | 36.75 |
| 7 | GLM-4 32B (0414) | 36.45 |
| 8 | Qwen 2.5 14B Instruct | 35.96 |
| 9 | Qwen 3 8B | 35.73 |
| 10 | Qwen 2.5 7B Instruct | 35.26 |
| 11 | Llama 3.3 70B Instruct | 34.34 |
| 12 | Llama 3.1 70B Instruct | 34.24 |
| 13 | GLM-4 9B (0414) | 33.12 |
| 14 | DeepSeek R1 0528 | 31.54 |
| 15 | Mistral 7B Instruct (v0.3) | 31.35 |
Interactive version: theaggregate.ai/benchmark?slug=communitybench-preference-identification · How It Works · Data refreshed daily, snapshot 2026-09-25.