BenCzechMark: leaderboard
Multitask Czech language benchmark evaluating LLMs across machine learning, chat, and familiarity tasks. 64+ models tested with statistical significance-based scoring.
Metric: Average Score (%). Source: huggingface.co. Status: saturation imminent. 74 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | DeepSeek V3 (0324) | 83.39 |
| 2 | Llama 3.1 405B Instruct | 83.19 |
| 3 | DeepSeek V4 Flash (0731) | 81.36 |
| 4 | DeepSeek R1 0528 | 81.06 |
| 5 | c4ai-command-a-03-2025 | 78.09 |
| 6 | Qwen 2.5 72B | 77.91 |
| 7 | Mistral Large 2 (Nov) Instruct (2411) | 74.24 |
| 8 | Gemma 4 31B (IT) | 72.45 |
| 9 | Llama 3.3 70B Instruct | 71.95 |
| 10 | MiniMax M1 80k | 71.78 |
| 11 | Qwen 2.5 72B Instruct | 71.26 |
| 12 | Qwen 3.6 27B | 70.57 |
| 13 | Llama 4 Scout Instruct | 69.79 |
| 14 | Llama 3.1 70B Instruct | 69.51 |
| 15 | Qwen 2.5 32B Instruct | 66.04 |
Interactive version: theaggregate.ai/benchmark?slug=benczechmark · How It Works · Data refreshed daily, snapshot 2026-09-05.