Swallow - Pre-trained Japanese - JCommonsenseQA: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) pre-trained Japanese board, which evaluates base models without post-training. JCommonsenseQA five-choice commonsense questions built from a knowledge base (JGLUE), 4-shot; metric is accuracy (random 20%). Upstream publishes 0-1 scores; shown as 0-100, higher is better.
Metric: Accuracy (%). Source: swallow-llm.github.io. Saturation forecast: Estimated already saturated. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 2.5 72B | 97.2 | #512 |
| 2 | Qwen 2.5 32B | 96.1 | #544 |
| 3 | Llama 4 Scout Base | 95.8 | #708 |
| 4 | Qwen 2.5 14B | 95.8 | #716 |
| 5 | Qwen3-14B-Base | 95.6 | #628 |
| 6 | Llama 3.1 70B | 94.6 | #588 |
| 7 | gemma-3-27B-pt | 94.4 | #505 |
| 8 | Gemma 2 27B | 93.6 | #803 |
| 9 | sarashina2-70B | 92.9 | #586 |
| 10 | Qwen3-8B-Base | 92.7 | #626 |
| 11 | Qwen3-30B-A3B-Base | 92.7 | #597 |
| 12 | Qwen 2.5 7B | 92.4 | #895 |
| 13 | sarashina2.2-3B | 91.1 | #766 |
| 14 | Qwen3-4B-Base | 91 | #880 |
| 15 | plamo-2-8B | 90.9 | #839 |
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-jcommonsenseqa · How It Works · Data refreshed daily, snapshot 2026-11-05.