MedPIC-Bench - Overall: leaderboard
Overall accuracy on MedPIC-Bench, 467 expert-validated medication-safety questions combining guideline-following cases with paired counterfactuals in which a controlled change in patient information changes whether a rule applies (MedPIC-Bench, arXiv 2608.03028v1, Table 2). Percent; higher is better; 28 medical, general and proprietary models.
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) | 80.7 | #64 |
| 2 | GPT-5 | 75.6 | #105 |
| 3 | Qwen 3.5 Plus | 72.4 | #133 |
| 4 | DeepSeek V4 Pro | 71.7 | #112 |
| 5 | GPT-OSS-120B | 70.7 | #346 |
| 6 | Qwen 3.5 27B | 69.4 | #192 |
| 7 | Qwen 3.5 35B A3B | 68.5 | #268 |
| 8 | GPT-5.2 | 68.1 | #115 |
| 9 | Claude Sonnet 4.6 | 64.2 | #94 |
| 10 | Qwen 3.5 9B | 63.8 | #391 |
| 11 | GPT-OSS-20B | 59.1 | #523 |
| 12 | Llama 3.1 70B | 55.7 | #588 |
| 13 | Gemini 2.5 Pro | 54.4 | #159 |
| 14 | Gemma 3 27B | 40.3 | #606 |
| 15 | Llama 3.1 8B | 40 | #1130 |
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-overall · How It Works · Data refreshed daily, snapshot 2026-11-05.