DeepSeek-R1-Distill-Qwen-32B-Japanese: benchmark results
CyberAgent's DeepSeek-R1-Distill-Qwen-32B-Japanese, a Japanese fine-tune of DeepSeek's R1-distilled Qwen 32B reasoning model. Provider: CyberAgent. Released 2025-01-27. Access: Open.
Aggregate Rating 1542 ± 17, rank #638 of 1656 rated models, from 186 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Nejumi 4 - jaster (2-shot) - MAWPS | 99 | Exact match (%) | 96.9 |
| Open Japanese LLM - Wiki NER SET F1 | 15.93 | Score (%) | 96.9 |
| Open Japanese LLM - Jsts Pearson | 90.31 | Score (%) | 89.7 |
| Open Japanese LLM - Jsem Exact Match | 80.24 | Score (%) | 89.5 |
| Open Japanese LLM - EL | 58.6 | Score (%) | 88.9 |
| Open Japanese LLM - ALT E TO J Comet Wmt22 | 91.01 | Score (%) | 88.8 |
| Open Japanese LLM - Jmmlu Exact Match | 72.38 | Score (%) | 80.8 |
| Open Japanese LLM - FA | 28.03 | Score (%) | 80 |
| Open Japanese LLM - Jsts Spearman | 87.55 | Score (%) | 79.8 |
| Open Japanese LLM - Xlsum JA Rouge1 | 29.71 | Score (%) | 79.2 |
| Open Japanese LLM - Jcommonsenseqa Exact Match | 95.62 | Score (%) | 79.1 |
| Open Japanese LLM - Xlsum JA Rouge2 | 10.75 | Score (%) | 77 |
Interactive version: theaggregate.ai/model?slug=deepseek-r1-distill-qwen-32b-japanese · How It Works · Data refreshed daily, snapshot 2026-10-20.