Open Japanese LLM - Xlsum JA Bleu JA: leaderboard
The "Xlsum JA Bleu JA" 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: Around 2034. 418 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Llama-3.1-Tulu-3-70B | 5.76 | #482 |
| 2 | aya-23-35B | 4.15 | #883 |
| 3 | Llama-3.3-Swallow-70B-Instruct-v0.4 | 3.77 | #529 |
| 4 | aya-expanse-32B | 3.66 | #692 |
| 5 | Llama 3.1 70B Instruct | 3.43 | #561 |
| 6 | Llama 3.1 Nemotron 70B Instruct | 3.41 | #554 |
| 7 | Llama 3.1 8B Instruct | 3.2 | #1035 |
| 8 | Qwentile2.5-32B-Instruct | 3.17 | #339 |
| 9 | granite-3.1-8B-instruct | 3.15 | #1069 |
| 10 | Llama 3.3 70B Instruct | 3.14 | #534 |
| 11 | QwQ 32B-Preview | 3.14 | #651 |
| 12 | Gemma 2 27B (IT) | 3.06 | #645 |
| 13 | tempmotacilla-cinerea-0308 | 2.93 | #395 |
| 14 | Lamarckvergence-14B | 2.9 | #545 |
| 15 | Qwen 2.5 14B Instruct | 2.84 | #647 |
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-xlsum-ja-bleu-ja · How It Works · Data refreshed daily, snapshot 2026-10-20.