855 free, open-source medical AI models, 2,269 downloadable builds, every one linked to its repository on Hugging Face. Built for physicians who need a job done and for developers who need a model that fits their stack. Updated 2026-09-17.
Start from what you need
De-identify patient text
Find and remove names, dates, identifiers and other personal data in 33 languages.
Drugs and chemicals
Medication and chemical mentions in notes and papers.
Diseases and conditions
Disease and disorder mentions for phenotyping and indexing.
Genes, proteins and cells
Genes, proteins, DNA, RNA, cell lines and cell types.
Cancer, anatomy and species
Cancer genetics, anatomical structures and organism names.
Zero-shot: your own entity labels
Give the model the labels you need at run time, no retraining.
Runs on phone, browser or Mac
ONNX builds for Android, browser and CPU; MLX builds for Apple Silicon.
Best model per benchmark
The top reported F1 on each of the benchmark corpora.
Vision-language and small LLMs
Models that read images and text, or generate medical text.
Explore all models
Search and filters are for registered members. Log in to search the 855 models by task, entity type, language, runtime, size and score. The model pages, the leaderboard and the language list below are open to everyone.
Models by language
The language a model reads. All descriptions on this site are in English.
- English 522
- German 40
- Spanish 40
- French 40
- Italian 40
- Hindi 39
- Dutch 39
- Portuguese 37
- Telugu 37
- Turkish 35
- Vietnamese 12
- Chinese 11
- Japanese 7
- Arabic 6
- Korean 6
- Bengali 4
- Polish 2
- Bulgarian 1
- Czech 1
- Danish 1
- Greek 1
- Estonian 1
- Finnish 1
- Croatian 1
- Hungarian 1
- Lithuanian 1
- Latvian 1
- Romanian 1
- Slovak 1
- Russian 1
- Persian 1
- Swahili 1
- Thai 1
Best model per benchmark
Highest F1 reported by the authors on each benchmark's test set. The 35 models whose cards state that no verified evaluation exists are not ranked here.
About this catalog
Models are harvested from the OpenMed organization on Hugging Face, grouped so that one card is one trained model and its PyTorch, ONNX and MLX builds are download options inside it. Scores, labels, sizes and base models are read from each repository; the short descriptions are written from those facts and reviewed automatically for accuracy. Read the OpenMed paper and the OpenMed library for the method behind the models. Nothing here is a medical device or a clinical validation; evaluate any model on your own data before use.
All model information shown here is public and belongs to its authors: use and download it freely under each model’s license. The selection, organization, search design and presentation of this catalog are © 2026 Clinical Corpus. Please link to these pages rather than copy them.