AI & ML interests
healthcare AI, clinical NER, biomedical NLP, PII / de-identification, on-device AI, edge inference, Apple Silicon / MLX, data sovereignty, privacy-preserving ML, open-source
Recent Activity
Articles
Health × AI
Open clinical NER & PII de-identification models — private, local-first, on-device.
Runs locally — in Python, on iOS and Android, in the browser, and on your own servers.
OpenMed ships a large, growing family of open medical-AI models: clinical named-entity recognition (diseases, chemicals, genes, anatomy, pharmacology, oncology) and privacy-preserving PII de-identification in 35 model-backed languages (42 supported language codes). The SDK and 2,200+ of the models are Apache-2.0, benchmark-driven, and small enough to run locally.
Start here
- Browse the models → the tabs above (2,200+ NER & PII models)
- Flagship clinical NER →
OpenMed-NER-PharmaDetect-SuperClinical-434M - Python →
pip install openmed - On-device → Swift OpenMedKit + MLX on Apple devices, ONNX on Android, Transformers.js in the browser
- Docs & guides → openmed.life/docs
Why OpenMed
| Private by design | Models run locally, so clinical text can stay on hardware you control. |
| State of the art | Clinical NER that set state of the art on 10 of 12 public biomedical NER benchmarks (arXiv 2508.01630). |
| Truly open | Apache-2.0 SDK, and Apache-2.0 weights and recipes for 2,200+ models. Free for research and production. |
| Runs anywhere | Python, iOS and macOS (MLX), Android (ONNX), the browser (Transformers.js), and servers. |
-
OpenMed/Qwen3-0.6B-WSP-Clinical
Text Generation • 0.8B • Updated • 497 -
OpenMed/Qwen3.5-0.8B-WSP-Clinical
Image-Text-to-Text • 0.9B • Updated • 46 -
OpenMed/Qwen3-1.7B-WSP-Clinical
Text Generation • 2B • Updated • 516 -
OpenMed clinical weight edit, side by side
🩺Qwen3.5-0.8B vs OpenMed's clinical weight edit
-
OpenMed/privacy-filter-nemotron-v2
Token Classification • 1B • Updated • 172k • • 57 -
OpenMed/privacy-filter-nemotron
Token Classification • 1B • Updated • 106k • 50 -
OpenMed/privacy-filter-multilingual-v2
Token Classification • 1B • Updated • 121k • 20 -
OpenMed/privacy-filter-multilingual
Token Classification • 1B • Updated • 118k • 24
-
OpenMed/Qwen3-0.6B-WSP-Clinical
Text Generation • 0.8B • Updated • 497 -
OpenMed/Qwen3.5-0.8B-WSP-Clinical
Image-Text-to-Text • 0.9B • Updated • 46 -
OpenMed/Qwen3-1.7B-WSP-Clinical
Text Generation • 2B • Updated • 516 -
OpenMed clinical weight edit, side by side
🩺Qwen3.5-0.8B vs OpenMed's clinical weight edit
-
OpenMed/privacy-filter-nemotron-v2
Token Classification • 1B • Updated • 172k • • 57 -
OpenMed/privacy-filter-nemotron
Token Classification • 1B • Updated • 106k • 50 -
OpenMed/privacy-filter-multilingual-v2
Token Classification • 1B • Updated • 121k • 20 -
OpenMed/privacy-filter-multilingual
Token Classification • 1B • Updated • 118k • 24
spaces 4
OpenMed clinical weight edit, side by side
Qwen3.5-0.8B vs OpenMed's clinical weight edit
OpenMed NER Model Discovery
It help you find the best medical and clinical NER models
⚕️ Openmed Clinical NER
Specialized in cancer, genetics, and oncology entities