MedGemma-27B-text-it: benchmark results
Google's MedGemma 27B text-it, a 27B-parameter medical-domain model in the MedGemma family, fine-tuned from Gemma 3 27B. Provider: Google. Released 2025-05-19. Access: Open.
Aggregate Rating 1576 ± 18, rank #1218 of 10870 rated models, from 78 benchmark results.
Strongest benchmark results
| Benchmark | Score | Metric | Percentile |
|---|---|---|---|
| MIMIC-CDM - Pancreatitis | 68.2 | Accuracy (%) | 100 |
| MedPIC-Bench - Rule Activation | 76.1 | Accuracy (%) | 100 |
| Open Portuguese LLM - ASSIN2 RTE | 94.85 | Macro F1 (%) | 99.1 |
| Open Portuguese LLM - FaQuAD NLI | 85.38 | Macro F1 (%) | 98.6 |
| Open Portuguese LLM - BLUEX | 75.38 | Accuracy (%) | 95.9 |
| HealthBench Hard | 57 | Overall score (imported) | 93.5 |
| MIMIC-CDM | 69.1 | Accuracy (%) | 92.9 |
| MIMIC-CDM - Appendicitis | 89.1 | Accuracy (%) | 92.9 |
| Open Portuguese LLM - ENEM | 79.78 | Accuracy (%) | 92.8 |
| Open Portuguese LLM - OAB Exams | 60.41 | Accuracy (%) | 92.7 |
| BRIDGE Medical Leaderboard - Few-Shot | 51.73 | Average Performance (%) | 90.1 |
| MIMIC-CDM - Diverticulitis | 61.1 | Accuracy (%) | 89.3 |
Interactive version: theaggregate.ai/model?slug=medgemma-27b-text-it · How It Works · Data refreshed daily, snapshot 2026-11-12.