KyrgyzLLM-Bench - KyrgyzMMLU: leaderboard
KyrgyzMMLU from KyrgyzLLM-Bench (OpenLLM Kyrgyz Leaderboard v0.1): native Kyrgyz multiple-choice knowledge questions, zero-shot. Accuracy 0-100, higher is better. Proprietary API models were run on a question subset while open-weight models used the full set; the small open models all score between 26 and 32.
Metric: Accuracy (%, zero-shot). Source: huggingface.co. Saturation forecast: Around April 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4.5 | 66.78 | #148 |
| 2 | DeepSeek V3 | 64 | #329 |
| 3 | Mistral Large 3 | 61.44 | #410 |
| 4 | GPT-5.1 (2025-11-13) | 59.56 | #107 |
| 5 | Qwen 3 Max | 59.56 | #213 |
| 6 | Claude Haiku 4.5 (20251001) | 59.11 | #264 |
| 7 | GigaChat 2 Max | 57.33 | #545 |
| 8 | Mistral Medium 3.1 | 53.56 | #418 |
| 9 | GPT-5 Mini (2025-08-07) | 50.11 | #175 |
| 10 | Gemini 2.5 Flash | 50 | #255 |
| 11 | Grok 4.1 Fast (Non-reasoning) | 48.56 | #227 (Grok 4.1 Fast) |
| 12 | Qwen 3 8B | 32.1 | #678 |
| 13 | Qwen 2.5 7B Instruct | 31.5 | #863 |
| 14 | Llama 3.1 8B Instruct | 31 | #1017 |
| 15 | Gemma 3 4B (IT) | 30.3 | #997 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=kyrgyzllm-bench-kyrgyzmmlu · How It Works · Data refreshed daily, snapshot 2026-11-05.