Medical Question Answering on MMLU-Med MIRAGE
88.6AccuracyMEDVRAG (full system)
Evaluation Results
| Method | Links | |
|---|---|---|
| MEDVRAG (full system)Memory Bank Usage=memory bank (full system)2026.04 | 88.6 | |
| MEDVRAG (+ iterative reasoning)Reasoning Rounds=iterative, Memory Bank Usage=no memory2026.04 | 88.4 | |
| MedRAG + GPT-4RAG Protocol=Text-only RAG2026.04 | 87.2 | |
| MEDVRAG (single-round)Reasoning Rounds=single-round, Data format=page images2026.04 | 86.8 | |
| MEDVRAG (text-only retrieval)Retrieval Type=text-only2026.04 | 86.2 | |
| Qwen3-30BRAG Protocol=No-RAG baselines2026.04 | 85 | |
| MEDVRAG (no RAG)RAG Protocol=No RAG2026.04 | 83.7 | |
| Yi-34B (OpenMedLM)RAG Protocol=No-RAG baselines2026.04 | 81.7 | |
| Qwen3-8BRAG Protocol=No-RAG baselines2026.04 | 78.8 | |
| Meditron-70BRAG Protocol=No-RAG baselines2026.04 | 77.6 | |
| MedRAG + GPT-3.5RAG Protocol=Text-only RAG2026.04 | 75.5 | |
| Self-BioRAG-7BRAG Protocol=Text-only RAG2026.04 | 53.9 |