BenCzechMark - Czech Language Understanding: leaderboard
Metric: Duel Win Score (%). Source: huggingface.co. 78 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V4 Flash (0731) | 92.42 |
| 2 | Llama 3.1 405B Instruct | 91.77 |
| 3 | DeepSeek V3 (0324) | 88.96 |
| 4 | Gemma 4 31B (IT) | 88.1 |
| 5 | DeepSeek R1 0528 | 86.58 |
| 6 | c4ai-command-a-03-2025 | 76.62 |
| 7 | Llama 4 Maverick Instruct | 76.41 |
| 8 | Gemma 4 26B A4B (IT) | 72.08 |
| 9 | Mistral Large 2 (Nov) Instruct (2411) | 72.08 |
| 10 | Llama 3.3 70B Instruct | 71.86 |
| 11 | Llama 3.1 70B Instruct | 67.75 |
| 12 | Qwen 2.5 72B Instruct | 67.32 |
| 13 | Llama 4 Scout Instruct | 67.1 |
| 14 | MiniMax M1 80k | 66.67 |
| 15 | Qwen 3.6 27B | 64.29 |
Interactive version: theaggregate.ai/benchmark?slug=benczechmark-czech-language-understanding · How It Works · Data refreshed daily, snapshot 2026-09-19.