Open Japanese LLM - MR: leaderboard
The "MR" subtask of Open Japanese LLM. A task from the Open Japanese LLM Leaderboard, evaluating models on Japanese-language knowledge, reasoning and understanding.
Metric: Score (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 418 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama 3 70B | 94.8 | #631 |
| 2 | Qwentile2.5-32B-Instruct | 94.6 | #339 |
| 3 | Llama 3.3 70B Instruct | 94.4 | #534 |
| 4 | Qwen 2.5 32B Instruct | 94.4 | #498 |
| 5 | Qwen2.5-32B-Instruct-CFT | 94.4 | #375 |
| 6 | Rombos-LLM-V2.5-Qwen-32B | 94.4 | #367 |
| 7 | Llama-3.3-Swallow-70B-Instruct-v0.4 | 94.2 | #529 |
| 8 | oxyge1-33B | 94.2 | #690 |
| 9 | Awqward2.5-32B-Instruct | 94 | #371 |
| 10 | ABEJA-Qwen2.5-32B-Japanese-v1.0 | 94 | #503 |
| 11 | QwQ 32B-Preview | 93.8 | #651 |
| 12 | qwen2.5-bakeneko-32B-instruct-v2 | 93.2 | #562 |
| 13 | Llama 3.1 70B Instruct | 93 | #561 |
| 14 | Llama 3.1 Nemotron 70B Instruct | 92.8 | #554 |
| 15 | EZO-Qwen2.5-32B-Instruct | 92.8 | #471 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=open-japanese-llm-mr · How It Works · Data refreshed daily, snapshot 2026-10-20.