Nejumi 4 - jaster (0-shot): leaderboard
The average of the per-dataset scores in the jaster battery when every task is asked with no worked examples. Jaster is a set of Japanese NLP datasets converted to one generation harness, covering question answering, inference, grammaticality, translation, math word problems and multiple-choice exams, each graded automatically. Reported in percent, 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 | GPT-5.5 (2026-04-23) | 88.02 | #22 |
| 2 | Claude Fable 5 | 87.82 | #16 |
| 3 | Gemini 3 Flash (Preview) | 87.61 | #82 |
| 4 | Gemini 3.1 Pro (Preview) | 87.57 | #58 |
| 5 | Gemini 3.6 Flash | 87.55 | #51 |
| 6 | Gemini 3.5 Flash | 87.47 | #60 |
| 7 | GPT-5.6 Terra | 87.04 | #41 |
| 8 | Gemini 3 Pro (Preview) | 86.77 | #66 |
| 9 | GPT-5.4 (2026-03-05) | 86.77 | #59 |
| 10 | GPT-5.6 Sol | 86.6 | #19 |
| 11 | Qwen3.8 2.4T A95B | 86.57 | #55 |
| 12 | Qwen 3.6 Max Preview | 86.13 | #85 |
| 13 | Claude Opus 5 | 85.94 | #12 |
| 14 | GPT-5.2 | 85.93 | #110 |
| 15 | Claude Opus 4.7 | 85.82 | #49 |
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 · How It Works · Data refreshed daily, snapshot 2026-10-20.