EHRStruct (Synthea) - Subtraction: leaderboard
Metric: Accuracy (%; D-R4: 100 zero-shot questions on synthetic Synthea patient tables given as plain text, where each item asks to subtract one numeric field of a patient from another (for example income minus healthcare expenses), rounded to two decimals; an answer counts when it matches the reference exactly; the paper's check mark, a perfect score, reads 100, and its cross, no valid output, reads as missing). Source: arxiv.org. Saturation forecast: Estimated already saturated. 16 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 2.5 Flash | 100 |
| 2 | DeepSeek V3 | 72 |
| 3 | GPT-4.1 | 70 |
| 4 | Gemini 2.0 Flash | 54 |
| 5 | DeepSeek V2.5 | 44 |
| 6 | GPT-3.5 Turbo | 7 |
Interactive version: theaggregate.ai/benchmark?slug=ehrstruct-synthea-subtraction · How It Works · Data refreshed daily, snapshot 2026-09-29.