MedRepBench - Reference Range Recall (OCR Text Only): leaderboard

Metric: Field-level recall (%; reference range of each ground-truth lab item, matched by exact normalized item name within the model's top-K items, K = number of reference items; text LLM on the OCR text). Source: arxiv.org. Saturation forecast: Estimated already saturated. 12 models tracked.

Top models

#ModelScore
1DeepSeek V392.6
2Qwen 3 235B A22B 250792.48
3Qwen 3 32B90.68
4Qwen 3 235B A22B90.56
5Qwen 3 14B90.18
6Qwen 3 30B A3B90.08
7DeepSeek R1 Distill Qwen 14B88.38
8Qwen 3 8B88.26
9DeepSeek R1 Distill Qwen 32B87.53
10Qwen 3 4B87.4
11Qwen 3 1.7B80.91
12DeepSeek-R1-Distill-Qwen-7B75.13

Interactive version: theaggregate.ai/benchmark?slug=medrepbench-reference-range-recall-ocr-text-only · How It Works · Data refreshed daily, snapshot 2026-09-26.