Nejumi 4 - MT-Bench (Japanese) - STEM: leaderboard
Japanese MT-Bench questions in the STEM category, two-turn answers rated 1-10 by GPT-4.1 as judge in Nejumi Leaderboard 4 (Weights & Biases Japan). Nejumi stores ratings divided by 10; the board multiplies by 100 (10-100), the scale of Nejumi 4 - MT-Bench (Japanese). Higher is better.
Metric: Judge rating (1-10, x10). Source: nejumi.ai. Saturation forecast: Estimated already saturated. 145 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 100 | #64 |
| 2 | Gemini 3 Pro (Preview) | 100 | #71 |
| 3 | Gemma 4 31B (IT) | 100 | #178 |
| 4 | Qwen 3 Next 80B A3B (Thinking) | 100 | #312 |
| 5 | Gemini 3.5 Flash | 100 | #66 |
| 6 | Qwen 3 235B A22B 2507 (Thinking) | 100 | #263 |
| 7 | DeepSeek V3.2 (Thinking) | 100 | #210 (DeepSeek V3.2) |
| 8 | GPT-5.6 Sol (Max) | 100 | #22 (GPT-5.6 Sol) |
| 9 | Kimi K3 (Max) | 100 | #35 (Kimi K3) |
| 10 | Kimi K2.5 (Thinking) | 100 | #152 (Kimi K2.5) |
| 11 | GLM-5.3 (Max) | 100 | #44 (GLM-5.3) |
| 12 | Claude Opus 5 (Max) | 100 | #16 (Claude Opus 5) |
| 13 | DeepSeek V4 Flash (Max) | 100 | #155 (DeepSeek V4 Flash) |
| 14 | Claude Opus 4.6 (Thinking) | 100 | #67 (Claude Opus 4.6) |
| 15 | GLM-5.3 Flash (Max) | 100 | #77 (GLM-5.3 Flash) |
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-mt-bench-japanese-stem · How It Works · Data refreshed daily, snapshot 2026-11-05.