Swallow - Post-trained Japanese - JHumanEval: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) post-trained Japanese board, which evaluates instruction-tuned and reasoning models with the Swallow evaluation harness. JHumanEval (Sato et al., 2024), a Japanese translation of HumanEval Python function synthesis, zero-shot; metric is pass@1 estimated 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. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (2026-03-05) (Medium) | 96.3 | #63 (GPT-5.4 (2026-03-05)) |
| 2 | gemma-4-26B-A4B-it (Thinking) | 95.9 | |
| 3 | Gemma 4 31B (IT) (Thinking) | 95.2 | #178 (Gemma 4 31B (IT)) |
| 4 | Qwen 3 235B A22B 2507 (Thinking) | 94.8 | #263 |
| 5 | GPT-5 (Medium) | 94.6 | #105 (GPT-5) |
| 6 | GPT-5 Mini (2025-08-07) (Medium) | 94.4 | #175 (GPT-5 Mini (2025-08-07)) |
| 7 | Qwen 3.5 122B A10B (Thinking) | 93.7 | #184 (Qwen 3.5 122B A10B) |
| 8 | O3 (2025-04-16) (Medium) | 93.7 | #125 (O3 (2025-04-16)) |
| 9 | GPT-OSS-Swallow-120B-RL-v0.1 (Thinking) | 93.4 | #309 (GPT-OSS-Swallow-120B-RL-v0.1) |
| 10 | GPT-OSS-20B (Medium) | 93.2 | #523 (GPT-OSS-20B) |
| 11 | Qwen 3.5 27B (Thinking) | 93 | #192 (Qwen 3.5 27B) |
| 12 | NVIDIA-Nemotron-3-Super-120B-A12B-BF16 (Thinking) | 93 | #394 (NVIDIA-Nemotron-3-Super-120B-A12B-BF16) |
| 13 | Qwen3-Swallow-32B-RL-v0.2 (Thinking) | 92.9 | #374 (Qwen3-Swallow-32B-RL-v0.2) |
| 14 | O3 Mini (Medium) | 92.8 | #282 (O3 Mini) |
| 15 | GPT-OSS-120B (Medium) | 92.7 | #346 (GPT-OSS-120B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swallow-post-trained-japanese-jhumaneval · How It Works · Data refreshed daily, snapshot 2026-11-05.