Nejumi 4 - GLP - Specialized Knowledge: leaderboard
Averages three expert knowledge sets: JMMLU and the Japanese MMLU-ProX multiple-choice exams run through the jaster harness at 0-shot and 2-shot, and a 194-question Japanese version of Humanity's Last Exam with multimodal items removed and inputs kept under 7000 tokens. All closed-book, reported in percent, higher is better. Source: github.com/wandb/llm-leaderboard.
Metric: Score (%). Source: nejumi.ai. Saturation forecast: Around January 2027. 139 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 | 81.39 | #16 |
| 2 | GPT-5.6 Sol | 79.19 | #19 |
| 3 | Claude Opus 5 | 78.84 | #12 |
| 4 | GPT-5.5 (2026-04-23) | 77.46 | #22 |
| 5 | Gemini 3.1 Pro (Preview) | 77.27 | #58 |
| 6 | GPT-5.6 Terra | 76.97 | #41 |
| 7 | Kimi K3 | 76.94 | #30 |
| 8 | GPT-5.4 (2026-03-05) | 76.63 | #59 |
| 9 | Hy4 preview | 76.28 | #38 |
| 10 | Claude Opus 4.8 | 75.99 | #37 |
| 11 | Grok 4.5 | 75.93 | #48 |
| 12 | Qwen3.8 2.4T A95B | 75.27 | #55 |
| 13 | DeepSeek V4.1 Flash | 75.1 | #36 |
| 14 | Claude Opus 4.7 | 74.94 | #49 |
| 15 | Muse Spark Open | 74.58 | #54 |
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-glp-specialized-knowledge · How It Works · Data refreshed daily, snapshot 2026-10-20.