MedStruct-S (Two-Shot) - Key-Conditioned QA: leaderboard
Extracting the value of a queried key from OCR-derived Chinese clinical report pages; MedStruct-S (Li, Wang and Qian, 2026). Exact-match accuracy in percent on questions whose gold value is not empty, two-shot prompting; higher is better.
Metric: Exact-match accuracy (%; value extracted for a queried key, non-null gold values only, OCR-derived Chinese clinical report pages, two-shot prompting). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 14B | 60.76 | #1235 |
| 2 | Qwen 2.5 32B Instruct | 55.93 | #1216 |
| 3 | Qwen 3 0.6B | 18.78 | #8392 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=medstruct-s-two-shot-key-conditioned-qa · How It Works · Data refreshed daily, snapshot 2026-11-12.