Nejumi 4 - GLP - Logical Reasoning: leaderboard
The reasoning category of the Japanese MT-Bench: ten open-ended reasoning prompts, each played out over two turns in free text. An LLM judge rates every answer from one to ten and the board rescales that average onto a 0 to 100 range where 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 | Inkling Small | 98.5 | #138 |
| 2 | Claude Fable 5 | 98 | #16 |
| 3 | Muse Spark Open | 97 | #54 |
| 4 | Inkling | 96.5 | #126 |
| 5 | Qwen3.8 2.4T A95B | 96.5 | #55 |
| 6 | GLM-5.3 Flash | 96 | #65 |
| 7 | Claude Opus 5 | 95.5 | #12 |
| 8 | Muse Glimmer 30B | 95 | #178 |
| 9 | Claude Sonnet 4.6 | 94.5 | #90 |
| 10 | MiniMax-M3 | 94.5 | #136 |
| 11 | Qwen 3 235B A22B 2507 (Thinking) | 94.5 | #260 (Qwen 3 235B A22B 2507) |
| 12 | Hy3 | 94.5 | #93 |
| 13 | Ornith-1.5-9B | 94.5 | |
| 14 | Gemini 3.1 Pro (Preview) | 94 | #58 |
| 15 | Qwen 3.5 4B | 94 | #475 |
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-logical-reasoning · How It Works · Data refreshed daily, snapshot 2026-10-20.