CPTU Bench: leaderboard
Polish language comprehension benchmark: evaluates LLMs on sentiment analysis, language understanding, phraseology, and tricky questions in Polish.
Metric: Average Score (1-5). Source: huggingface.co. Saturation forecast: Around January 2027. 102 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 27B (Reasoning) | 4.41 | #170 (Qwen 3.6 27B) |
| 2 | Qwen 3.5 122B A10B (Reasoning) | 4.36 | #173 (Qwen 3.5 122B A10B) |
| 3 | Qwen 3.5 27B (Thinking) | 4.34 | #187 (Qwen 3.5 27B) |
| 4 | Gemma 4 31B (IT) (Non-reasoning) | 4.32 | #176 (Gemma 4 31B (IT)) |
| 5 | Gemini 2.0 Flash (001) | 4.29 | #360 |
| 6 | Qwen 3.6 27B (Non-reasoning) | 4.29 | #170 (Qwen 3.6 27B) |
| 7 | Qwen 3.5 27B (Non-reasoning) | 4.27 | #187 (Qwen 3.5 27B) |
| 8 | Qwen 3.5 122B A10B (Non-reasoning) | 4.26 | #173 (Qwen 3.5 122B A10B) |
| 9 | Qwen 3.5 35B A3B (Thinking) | 4.22 | #256 (Qwen 3.5 35B A3B) |
| 10 | Gemma 4 26B A4B (IT) (Reasoning) | 4.21 | #297 (Gemma 4 26B A4B (IT)) |
| 11 | Qwen 3.5 35B A3B (Non-reasoning) | 4.18 | #256 (Qwen 3.5 35B A3B) |
| 12 | DeepSeek V3.2 | 4.14 | #205 |
| 13 | DeepSeek R1 | 4.14 | #251 |
| 14 | Gemini 2.0 Flash Lite (001) | 4.09 | #544 |
| 15 | DeepSeek V3.1 | 4.03 | #264 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=cptu-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.