Nejumi 4 - jaster (0-shot) - JSQuAD: leaderboard
JSQuAD (JGLUE), Japanese extractive reading comprehension, from the llm-jp-eval jaster suite as run 0-shot by Nejumi Leaderboard 4 (Weights & Biases Japan). Score is character-level F1 x100 on Nejumi's sampled test split; higher is better.
Metric: Character F1 (x100). Source: nejumi.ai. Saturation forecast: Around May 2027. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Hy4 preview (High) | 83.3 | #45 (Hy4 preview) |
| 2 | Qwen 3.6 27B (Thinking) | 83 | #182 (Qwen 3.6 27B) |
| 3 | Gemini 3.6 Flash | 82.76 | #57 |
| 4 | Gemma 4 31B (IT) | 82.48 | #178 |
| 5 | Claude Opus 5 (Max) | 82.38 | #16 (Claude Opus 5) |
| 6 | ELYZA-Shortcut-1.0-Qwen-32B | 82.25 | #560 |
| 7 | Gemini 3 Pro (Preview) | 82.23 | #71 |
| 8 | DeepSeek V4.1 Flash (1017) | 82.12 | #27 |
| 9 | Gemini 3.5 Flash | 81.97 | #66 |
| 10 | Qwen 3 Next 80B A3B Instruct | 81.97 | #381 |
| 11 | mistral-large-4-oss (High) | 81.56 | #60 (mistral-large-4-oss) |
| 12 | MiMo-V2.5-Pro (Thinking) | 81.23 | #167 (MiMo-V2.5-Pro) |
| 13 | Kimi K3 (Thinking) | 81.22 | #35 (Kimi K3) |
| 14 | Solar Pro 4 | 81.19 | #230 |
| 15 | GPT-5.4 Mini (2026-03-17) (High) | 81.18 | #180 (GPT-5.4 Mini (2026-03-17)) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=nejumi-4-jaster-0-shot-jsquad · How It Works · Data refreshed daily, snapshot 2026-11-05.