BenCzechMark - NLI: leaderboard
BenCzechMark NLI category of Czech tasks. Scores are duel win scores (0-100): the share of pairwise statistical duels a model wins against every other model on the category's tasks, averaged over tasks, so they shift as the roster changes; higher is better.
Metric: Duel Win Score (%). Source: huggingface.co. Saturation forecast: Around February 2027. 79 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3.1 405B Instruct | 81.89 | #470 |
| 2 | DeepSeek V4 Flash (0731) | 79.49 | #84 |
| 3 | Qwen 2.5 72B Instruct | 79.49 | #469 |
| 4 | Gemma 4 31B (IT) | 77.4 | #178 |
| 5 | Qwen 2.5 32B Instruct | 77.16 | #509 |
| 6 | Qwen 3.6 27B | 76.92 | #182 |
| 7 | Qwen 2.5 72B | 76.76 | #512 |
| 8 | Llama 3.1 70B Instruct | 75.48 | #567 |
| 9 | c4ai-command-a-03-2025 | 75.32 | #468 |
| 10 | Mistral Large 2 (Nov) Instruct (2411) | 75.24 | #480 |
| 11 | Llama 3.3 70B Instruct | 74.44 | #541 |
| 12 | DeepSeek R1 0528 | 74.44 | #232 |
| 13 | DeepSeek V3 (0324) | 73.56 | #349 |
| 14 | Gemma 3 27B (IT) | 72.76 | #528 |
| 15 | Gemma 4 26B A4B (IT) | 71.63 | #293 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=benczechmark-nli · How It Works · Data refreshed daily, snapshot 2026-11-05.