BenCzechMark: leaderboard
Czech benchmark scored by statistically significant task-duel wins against the current submitted roster. The overall score is the equal-weight mean of seven category win percentages; language-modeling tasks are excluded by default.
Metric: Average Duel Win Score (%). Source: huggingface.co. Saturation forecast: Around January 2027. 78 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | DeepSeek V3 (0324) | 83.12 | #340 |
| 2 | Llama 3.1 405B Instruct | 82.65 | #457 |
| 3 | DeepSeek V4 Flash (0731) | 81.3 | #78 |
| 4 | DeepSeek R1 0528 | 80.84 | #222 |
| 5 | Qwen 2.5 72B | 77.48 | #499 |
| 6 | c4ai-command-a-03-2025 | 77.39 | #451 |
| 7 | Mistral Large 2 (Nov) Instruct (2411) | 73.17 | #465 |
| 8 | Llama 4 Maverick Instruct | 72.92 | #456 |
| 9 | Gemma 4 31B (IT) | 71.99 | #176 |
| 10 | Llama 3.3 70B Instruct | 71.29 | #534 |
| 11 | MiniMax M1 80k | 70.71 | #432 |
| 12 | Qwen 2.5 72B Instruct | 70.32 | #454 |
| 13 | Qwen 3.6 27B | 70.08 | #170 |
| 14 | Llama 4 Scout Instruct | 68.87 | #539 |
| 15 | Llama 3.1 70B Instruct | 68.77 | #561 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=benczechmark · How It Works · Data refreshed daily, snapshot 2026-10-20.