ABEJA-Qwen2.5-32B-Japanese-v1.0: benchmark results
ABEJA's Qwen2.5-32B Japanese v1.0, continually pretrained from Qwen2.5-32B-Instruct with a Japanese-centric corpus, then post-trained with SFT and DPO from its own v0.1 checkpoint. Provider: ABEJA. Released 2025-05-14. Access: Open.
Aggregate Rating 1582 ± 22, rank #503 of 1656 rated models, from 152 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| Open Japanese LLM - Jcommonsenseqa Exact Match | 97.14 | Score (%) | 99.8 |
| Open Japanese LLM - Jmmlu Exact Match | 77.69 | Score (%) | 99.3 |
| Open Japanese LLM - MC | 91.7 | Score (%) | 98.3 |
| Open Japanese LLM - HE | 78.81 | Score (%) | 98.1 |
| Nejumi 4 - JHumanEval | 55.13 | Sandbox pass rate (%) | 97.2 |
| Open Japanese LLM - Kuci Exact Match | 85.29 | Score (%) | 97.1 |
| Open Japanese LLM - Wiki PAS SET F1 | 11.64 | Score (%) | 96 |
| Open Japanese LLM - Jemhopqa Char F1 | 65.36 | Score (%) | 95.9 |
| Open Japanese LLM - NLI | 81.46 | Score (%) | 95 |
| Open Japanese LLM - Commonsensemoralja Exact Match | 92.69 | Score (%) | 94.2 |
| Open Japanese LLM - Jsts Pearson | 90.42 | Score (%) | 92.5 |
| Open Japanese LLM - Jnli Exact Match | 89.52 | Score (%) | 92.3 |
Interactive version: theaggregate.ai/model?slug=abeja-qwen2-5-32b-japanese-v1-0 · How It Works · Data refreshed daily, snapshot 2026-10-20.