List your company

Medical Question Answering on MMLU-Med MIRAGE

88.6AccuracyMEDVRAG (full system)
52.51261.88171.2580.619Apr 30, 2026

Evaluation results

MethodLinks
MEDVRAG (full system)Memory Bank Usage=memory bank (full system)2026.0488.6
MEDVRAG (+ iterative reasoning)Reasoning Rounds=iterative, Memory Bank Usage=no memory2026.0488.4
MedRAG + GPT-4RAG Protocol=Text-only RAG2026.0487.2
MEDVRAG (single-round)Reasoning Rounds=single-round, Data format=page images2026.0486.8
MEDVRAG (text-only retrieval)Retrieval Type=text-only2026.0486.2
Qwen3-30BRAG Protocol=No-RAG baselines2026.0485
MEDVRAG (no RAG)RAG Protocol=No RAG2026.0483.7
Yi-34B (OpenMedLM)RAG Protocol=No-RAG baselines2026.0481.7
Qwen3-8BRAG Protocol=No-RAG baselines2026.0478.8
Meditron-70BRAG Protocol=No-RAG baselines2026.0477.6
MedRAG + GPT-3.5RAG Protocol=Text-only RAG2026.0475.5
Self-BioRAG-7BRAG Protocol=Text-only RAG2026.0453.9