TMMLU+: leaderboard
TMMLU+ is a Traditional Chinese multi-subject exam benchmark (ikala): 66 subjects drawn from Taiwanese professional, high-school and university examinations, from accounting and pharmacy to Taiwanese Hokkien, answered as multiple choice. Only models that have completed all 66 subjects appear on the card's board.
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 96.16 | #16 |
| 2 | GPT-5.6 Sol | 88.38 | #22 |
| 3 | Claude Fable 5 | 87.06 | #20 |
| 4 | DeepSeek V4 Pro (0813) | 86.42 | #58 |
| 5 | Claude Sonnet 5 | 84.66 | #65 |
| 6 | Gemini 3.1 Pro (Preview) (Thinking) | 84.26 | #64 (Gemini 3.1 Pro (Preview)) |
| 7 | DeepSeek V4 Flash | 83.5 | #155 |
| 8 | Hy3 | 83.33 | #100 |
| 9 | GPT-5.6 Terra | 82.43 | #40 |
| 10 | Kimi K3 | 81 | #35 |
| 11 | Qwen 3.7 Max | 80.59 | #85 |
| 12 | Qwen 3.8 27B | 80.05 | #123 |
| 13 | GPT-5.6 Luna | 79.63 | #79 |
| 14 | GLM-5.2 | 78.59 | #80 |
| 15 | MiMo-V2.5 | 78.12 | #196 |
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 · How It Works · Data refreshed daily, snapshot 2026-11-05.