MedRepBench - Value Recall (OCR Text Only): leaderboard

Metric: Field-level recall (%; measured value 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
1Qwen 3 235B A22B 250789.4
2DeepSeek V388.83
3Qwen 3 32B87.46
4Qwen 3 14B87.43
5Qwen 3 235B A22B87.27
6Qwen 3 30B A3B86.48
7DeepSeek R1 Distill Qwen 14B84.61
8DeepSeek R1 Distill Qwen 32B83.4
9Qwen 3 8B79.48
10Qwen 3 4B74.59
11DeepSeek-R1-Distill-Qwen-7B70.6
12Qwen 3 1.7B70.5

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