TMMLU+ - STEM: leaderboard
Metric: Accuracy (%). Source: huggingface.co. 21 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 5 | 98.71 |
| 2 | GPT-5.6 Sol | 93.28 |
| 3 | Qwen 3.8 27B | 93.1 |
| 4 | DeepSeek V4 Pro (0813) | 92.89 |
| 5 | Claude Fable 5 | 90.61 |
| 6 | GPT-5.6 Terra | 90.22 |
| 7 | Hy3 | 89.81 |
| 8 | DeepSeek V4 Flash | 89.42 |
| 9 | Claude Sonnet 5 | 89.21 |
| 10 | GPT-5.6 Luna | 87.59 |
| 11 | MiMo-V2.5 | 87.44 |
| 12 | Gemini 3.1 Pro (Preview) (Thinking) | 87.14 |
| 13 | MiniMax-M3 | 86.21 |
| 14 | Kimi K3 | 86.12 |
| 15 | GLM-5.2 | 85.31 |
Interactive version: theaggregate.ai/benchmark?slug=tmmlu-plus-stem · How It Works · Data refreshed daily, snapshot 2026-09-19.