RxEval - Exact Match: leaderboard

Exact-match accuracy (%) on RxEval (2026), a prescription-level benchmark of 1,547 multiple-answer questions from MIMIC-IV inpatient trajectories: the model must select exactly the set of medication, dose and route triples that were prescribed at the time point, among distractors made by perturbing the clinical reasoning chain. Higher is better.

Metric: Exact match (%): the selected option set equals the set of medication-dose-route triples actually prescribed, over 1,547 RxEval multiple-answer questions, zero-shot at temperature 0; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 16 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1GPT-548.7#178
2GPT-5.447.9#135
3Gemini 3.1 Pro (Preview)46.1#99
4GPT-5 Mini42.7#364
5Qwen 3.6 27B40#351
6Gemini 2.5 Flash38#463
7Gemini 2.5 Pro37.6#276
8Gemini 3 Flash36.1#175
9GPT-4o21.7#751
10RxEval MedGemma-27B (checkpoint unspecified)16.7
11HuatuoGPT-o1-72B10.3#986
12Qwen 2.5 7B Instruct10.1#2723
13Qwen 2.5 72B Instruct9.9#1117
14Gemma 3 27B (IT)9.8#1324
15HuatuoGPT-o1-8B2.71#2983

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=rxeval-exact-match · How It Works · Data refreshed daily, snapshot 2026-11-21.