Swallow - Pre-trained Japanese - JSQuAD: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) pre-trained Japanese board, which evaluates base models without post-training. JSQuAD extractive reading comprehension over Japanese Wikipedia (JGLUE), 4-shot; metric is character-level F1. Upstream publishes 0-1 scores; shown as 0-100, higher is better.
Metric: Character F1 (%). Source: swallow-llm.github.io. Saturation forecast: Estimated already saturated. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B | 93 | #512 |
| 2 | sarashina2-70B | 92.9 | #586 |
| 3 | Llama 3.1 70B | 92.5 | #588 |
| 4 | Qwen 2.5 32B | 92.5 | #544 |
| 5 | Qwen 2.5 14B | 92.3 | #716 |
| 6 | Qwen3-14B-Base | 92.1 | #628 |
| 7 | Qwen3-30B-A3B-Base | 91.8 | #597 |
| 8 | Gemma 2 27B | 91.6 | #803 |
| 9 | Llama 4 Scout Base | 91.5 | #708 |
| 10 | gemma-3-27B-pt | 91.5 | #505 |
| 11 | Qwen3-8B-Base | 91.2 | #626 |
| 12 | gemma-3-12B-pt | 91.1 | #686 |
| 13 | plamo-2-8B | 91 | #839 |
| 14 | Qwen3-4B-Base | 90.8 | #880 |
| 15 | Qwen 2.5 7B | 90.7 | #895 |
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-jsquad · How It Works · Data refreshed daily, snapshot 2026-11-05.