MedPIC-Bench - Guideline-Following: leaderboard
Accuracy on MedPIC-Bench guideline-following questions, which test whether a model applies the relevant source-verifiable medication-safety recommendation to a fixed patient case (MedPIC-Bench, arXiv 2608.03028v1, Table 2). Percent; higher is better.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Around February 2027. 28 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 87.7 | #64 |
| 2 | GPT-5 | 83.5 | #105 |
| 3 | Qwen 3.5 Plus | 81.7 | #133 |
| 4 | DeepSeek V4 Pro | 81.3 | #112 |
| 5 | GPT-OSS-120B | 79.6 | #346 |
| 6 | GPT-5.2 | 78.2 | #115 |
| 7 | Qwen 3.5 27B | 78.2 | #192 |
| 8 | Qwen 3.5 35B A3B | 77.5 | #268 |
| 9 | Claude Sonnet 4.6 | 74.6 | #94 |
| 10 | Qwen 3.5 9B | 73.9 | #391 |
| 11 | GPT-OSS-20B | 66.5 | #523 |
| 12 | Llama 3.1 70B | 61.3 | #588 |
| 13 | Gemini 2.5 Pro | 60.6 | #159 |
| 14 | MedGemma-4B | 45.4 | #736 |
| 15 | Gemma 3 27B | 44.7 | #606 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=medpic-bench-guideline-following · How It Works · Data refreshed daily, snapshot 2026-11-05.