TreeProbe (Zero-Shot): leaderboard
TreeProbe (UESTC and Tibet University, 2026) evaluates large language models on Tibetan-medicine multiple-choice questions organized around the native Tree of Medicine framework, over ten subtasks in three diagnostic roots. Accuracy is the unweighted macro mean over the subtasks in percent, higher is better, under zero-shot prompting with English instructions and Tibetan-script content.
Metric: MCQ accuracy (%; unweighted macro mean over the ten subtasks; over TreeProbe's 3,289 expert-adjudicated Tibetan-medicine multiple-choice items in ten subtasks across three diagnostic roots, option order randomized, English prompt with Tibetan-script content; zero-shot prompting). Source: arxiv.org. Saturation forecast: Around February 2027. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 397B A17B | 60.55 | #242 |
| 2 | Gemini 3.1 Flash Lite | 59.89 | #363 |
| 3 | GPT-5.4 | 58.1 | #121 |
| 4 | GLM-4.7 | 52.11 | #365 |
| 5 | DeepSeek V3.2 (Non-reasoning) | 42.5 | #403 (DeepSeek V3.2) |
| 6 | Claude Sonnet 4.6 | 40.61 | #158 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=treeprobe-zero-shot · How It Works · Data refreshed daily, snapshot 2026-11-12.