Swallow - Pre-trained Japanese - WMT20 En-Ja: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) pre-trained Japanese board, which evaluates base models without post-training. English-to-Japanese news translation from WMT20, 4-shot, scored with corpus BLEU (0-100, higher is better).
Metric: BLEU. Source: swallow-llm.github.io. Saturation forecast: Estimated already saturated. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | sarashina2-70B | 31.3 | #586 |
| 2 | gemma-3-27B-pt | 30.1 | #505 |
| 3 | Llama 4 Scout Base | 30 | #708 |
| 4 | Gemma 2 27B | 29.5 | #803 |
| 5 | gemma-3-12B-pt | 28.8 | #686 |
| 6 | Llama 3.1 70B | 28.7 | #588 |
| 7 | Qwen 2.5 72B | 28.7 | #512 |
| 8 | sarashina2-13B | 28.4 | #898 |
| 9 | plamo-2-8B | 28 | #839 |
| 10 | llm-jp-3-13B | 27.3 | #1176 |
| 11 | sarashina2.2-3B | 27.3 | #766 |
| 12 | sarashina2-7B | 27.3 | #1134 |
| 13 | Qwen 2.5 32B | 27.1 | #544 |
| 14 | Gemma 2 9B | 26.9 | #924 |
| 15 | Qwen 2.5 14B | 26 | #716 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swallow-pre-trained-japanese-wmt20-en-ja · How It Works · Data refreshed daily, snapshot 2026-11-05.