Nejumi 4 - GLP - Expression: leaderboard
The roleplay, writing and humanities questions of the Japanese MT-Bench, ten open-ended prompts in each category, every one played out over two turns. An LLM judge rates each answer from one to ten, and the board averages the three category scores onto a 0 to 100 scale 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 | Qwen 3.6 Max Preview | 99.5 | #85 |
| 2 | Claude Sonnet 4.6 | 99.33 | #90 |
| 3 | Kimi K3 | 99.33 | #30 |
| 4 | Inkling | 99.33 | #126 |
| 5 | Gemini 3.6 Flash | 99.17 | #51 |
| 6 | Claude Opus 4.6 | 99 | #63 |
| 7 | Claude Opus 4.7 | 99 | #49 |
| 8 | Nex-N2.5-Pro | 99 | |
| 9 | Gemini 3.5 Flash | 98.83 | #60 |
| 10 | MiniMax-M3 | 98.83 | #136 |
| 11 | Gemini 3.1 Flash Lite (Preview) | 98.83 | #194 |
| 12 | Claude Opus 5 | 98.83 | #12 |
| 13 | Claude Opus 4.8 | 98.67 | #37 |
| 14 | GLM-5.3 | 98.67 | #39 |
| 15 | Kimi K2.5 | 98.5 | #144 |
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-expression · How It Works · Data refreshed daily, snapshot 2026-10-20.