TMMLU+ - STEM: leaderboard
The "STEM" category of TMMLU+, the Traditional Chinese multi-subject exam benchmark (ikala): 66 subjects drawn from Taiwanese professional, high-school and university examinations, answered as multiple choice and reported per category as well as overall.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 98.71 | #16 |
| 2 | GPT-5.6 Sol | 93.28 | #22 |
| 3 | Qwen 3.8 27B | 93.1 | #123 |
| 4 | DeepSeek V4 Pro (0813) | 92.89 | #58 |
| 5 | Claude Fable 5 | 90.61 | #20 |
| 6 | GPT-5.6 Terra | 90.22 | #40 |
| 7 | Hy3 | 89.81 | #100 |
| 8 | DeepSeek V4 Flash | 89.42 | #155 |
| 9 | Claude Sonnet 5 | 89.21 | #65 |
| 10 | GPT-5.6 Luna | 87.59 | #79 |
| 11 | MiMo-V2.5 | 87.44 | #196 |
| 12 | Gemini 3.1 Pro (Preview) (Thinking) | 87.14 | #64 (Gemini 3.1 Pro (Preview)) |
| 13 | MiniMax-M3 | 86.21 | #151 |
| 14 | Kimi K3 | 86.12 | #35 |
| 15 | GLM-5.2 | 85.31 | #80 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=tmmlu-plus-stem · How It Works · Data refreshed daily, snapshot 2026-11-05.