Swallow - Pre-trained Japanese - JHumanEval: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) pre-trained Japanese board, which evaluates base models without post-training. JHumanEval Japanese-prompt Python function synthesis, 0-shot; metric is pass@1 from 10 samples. Upstream publishes 0-1 scores; shown as 0-100, higher is better.
Metric: Pass@1 (%). Source: swallow-llm.github.io. Saturation forecast: Estimated already saturated. 50 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3-14B-Base | 70.9 | #628 |
| 2 | Qwen 2.5 72B | 64.8 | #512 |
| 3 | Qwen3-30B-A3B-Base | 64.1 | #597 |
| 4 | Qwen 2.5 32B | 63.7 | #544 |
| 5 | Qwen3-8B-Base | 59.5 | #626 |
| 6 | Qwen 2.5 14B | 55 | #716 |
| 7 | Qwen3-4B-Base | 53.7 | #880 |
| 8 | Qwen 2.5 7B | 50.7 | #895 |
| 9 | Gemma 2 27B | 49 | #803 |
| 10 | gemma-3-27B-pt | 47.3 | #505 |
| 11 | Llama 3.1 70B | 46.2 | #588 |
| 12 | Falcon3-10B-Base | 42.6 | #1187 |
| 13 | Qwen 2.5 3B | 40.4 | #1177 |
| 14 | gemma-3-12B-pt | 38.5 | #686 |
| 15 | Falcon3-7B-Base | 36.1 | #1286 |
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-jhumaneval · How It Works · Data refreshed daily, snapshot 2026-11-05.