Nejumi 4 - jaster (2-shot): leaderboard
The same jaster battery of Japanese NLP datasets, run with two worked examples in front of each question and averaged over its per-dataset scores. Nejumi 4 categories built from these per-dataset scores use the mean of this run and the 0-shot run rather than either alone, except ethics and morality, which reads this run only. Reported in percent where higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 88.63 | #16 |
| 2 | Gemini 3.1 Pro (Preview) | 88.33 | #58 |
| 3 | GPT-5.5 (2026-04-23) | 88.27 | #22 |
| 4 | GPT-5.6 Sol | 88.23 | #19 |
| 5 | Gemini 3 Flash (Preview) | 88.07 | #82 |
| 6 | Gemini 3.5 Flash | 87.94 | #60 |
| 7 | Claude Opus 5 | 87.92 | #12 |
| 8 | Gemini 3.6 Flash | 87.73 | #51 |
| 9 | Gemini 3 Pro (Preview) | 87.49 | #66 |
| 10 | GPT-5.6 Terra | 87.49 | #41 |
| 11 | Claude Opus 4.7 | 87.4 | #49 |
| 12 | GPT-5.4 (2026-03-05) | 87.16 | #59 |
| 13 | Claude Opus 4.8 | 86.97 | #37 |
| 14 | Kimi K3 | 86.92 | #30 |
| 15 | Qwen3.8 2.4T A95B | 86.79 | #55 |
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-2-shot · How It Works · Data refreshed daily, snapshot 2026-10-20.