ObGynLongBench - Visit-Level EHR: leaderboard
Metric: Accuracy (%; same-day visit records). Source: arxiv.org. 17 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3 Flash | 67.1 |
| 2 | GPT-5.4 Mini | 62.3 |
| 3 | DeepSeek V4 Flash | 61.2 |
| 4 | Lingshu-7B | 60.4 |
| 5 | Claude Haiku 4.5 | 58.5 |
| 6 | Qwen 3.5 35B A3B (Non-reasoning) | 55.7 |
| 7 | Qwen 3.5 9B (Non-reasoning) | 55.4 |
| 8 | Qwen 3 VL 4B | 54.7 |
| 9 | Qwen 3.5 4B (Non-reasoning) | 52.7 |
| 10 | Phi-4 Mini Instruct | 48.6 |
| 11 | MedGemma 1.5 4B | 48.1 |
| 12 | Gemma 4 E4B (Non-reasoning) | 47.4 |
Interactive version: theaggregate.ai/benchmark?slug=obgynlongbench-visit-level-ehr · How It Works · Data refreshed daily, snapshot 2026-09-19.