Swallow - Pre-trained Japanese - WMT20 Ja-En: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) pre-trained Japanese board, which evaluates base models without post-training. Japanese-to-English 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 | Llama 4 Scout Base | 25.8 | #708 |
| 2 | Llama 3.1 70B | 25.7 | #588 |
| 3 | gemma-3-27B-pt | 25.5 | #505 |
| 4 | Qwen 2.5 72B | 25.2 | #512 |
| 5 | Gemma 2 27B | 25.1 | #803 |
| 6 | gemma-3-12B-pt | 24.4 | #686 |
| 7 | sarashina2-70B | 24.3 | #586 |
| 8 | Gemma 2 9B | 23.6 | #924 |
| 9 | Qwen 2.5 32B | 23.3 | #544 |
| 10 | Qwen 2.5 14B | 23 | #716 |
| 11 | Qwen3-14B-Base | 22.9 | #628 |
| 12 | Qwen3-30B-A3B-Base | 22.4 | #597 |
| 13 | sarashina2-13B | 22.1 | #898 |
| 14 | Qwen3-8B-Base | 21.5 | #626 |
| 15 | Llama 3.1 8B | 21 | #1130 |
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-ja-en · How It Works · Data refreshed daily, snapshot 2026-11-05.