{"architecture":{"backbone":"ViT-g/14","family":"Vision Transformer","input":"224\u00d7224 tile at 20\u00d7","output":"1,536-dim embedding","params":"1.1B","pretraining":"self-supervised (DINOv2 family)"},"article":null,"cancer_slugs":["pan-cancer"],"category":"pathology","confidence":"medium","datasets":[{"name":"CAMELYON16 / CAMELYON17","note":"","role":"evaluation","slug":"camelyon16"},{"name":"TCGA \u2014 The Cancer Genome Atlas (via NCI Genomic Data Commons)","note":"","role":"evaluation","slug":"tcga"}],"developer":"Bioptimus","evaluation":[{"benchmark":"Camelyon16 \u2014 breast cancer detection (slide level, ABMIL)","external":false,"metric":"AUC-ROC","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.989 (SD 0.003); developer-run benchmark, average over 50 trainings"},{"benchmark":"PAIP 2020 \u2014 MSI prediction in colorectal cancer (slide level, ABMIL)","external":false,"metric":"AUC-ROC","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.967 (SD 0.007); developer-run benchmark, average over 50 trainings"},{"benchmark":"SLN-Breast \u2014 breast cancer detection (slide level, ABMIL)","external":false,"metric":"AUC-ROC","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.934 (SD 0.006); developer-run benchmark, average over 50 trainings"},{"benchmark":"TCGA-STAD-Kather \u2014 MSI prediction in gastric cancer (slide level, ABMIL)","external":false,"metric":"AUC-ROC","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.846 (SD 0.004); developer-run benchmark, average over 50 trainings"},{"benchmark":"Yale HER2 \u2014 HER2 status prediction in breast cancer (slide level, ABMIL)","external":false,"metric":"AUC-ROC","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.828 (SD 0.009); developer-run benchmark, average over 50 trainings"},{"benchmark":"CAMELYON17-WILDS, CRC-100K (normalised / not normalised), MHIST, TCGA-UNIFORM \u2014 tile classification (linear probing)","external":false,"metric":"accuracy","source":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0","value":"0.982 / 0.962 / 0.956 / 0.828 / 0.843 (SD 0.001-0.006); developer-run benchmark, linear classifier on frozen features, mean of three trainings"}],"hf":{"downloads":80579,"fetched_at":"2026-09-09T21:33:09Z","gated":"auto","last_modified":"2025-12-12","library":"timm","license":"apache-2.0","likes":92,"pipeline_tag":"image-feature-extraction"},"kind":"foundation","license":"apache-2.0","limitations":"- No peer-reviewed paper at release; details of the pretraining cohort are only partially disclosed.\n- Large model: slower tile throughput than ViT-L encoders.","links":{"demo":null,"docs":null,"doi":null,"github":null,"huggingface":"https://huggingface.co/bioptimus/H-optimus-0","paper":null,"pmid":null},"modalities":["histopathology"],"name":"H-optimus-0","notable_uses":"","openness":"gated-weights","regulatory":{"intended_use_en":"Research; not a medical device.","intended_use_pl":"Badania; nie jest wyrobem medycznym.","source_url":"https://huggingface.co/bioptimus/H-optimus-0","status":"research-only"},"regulatory_status":"research-only","release_date":"2024-07-10","run_snippet":"# generic timm loader \u2014 check the model card for the exact init args and image normalisation\nimport timm, torch\nfrom huggingface_hub import login\nlogin()  # gated repos: accept the licence on huggingface.co first\nmodel = timm.create_model('hf-hub:bioptimus/H-optimus-0', pretrained=True)\nmodel.eval()\ncfg = timm.data.resolve_data_config({}, model=model)\ntransform = timm.data.create_transform(**cfg)\n# emb = model(transform(tile).unsqueeze(0))  # 1 x D tile embedding\n","settings":["basic-research","diagnosis","prognosis"],"slug":"h-optimus-0","sources":[{"label":"Hugging Face model card \u2014 bioptimus/H-optimus-0","url":"https://huggingface.co/bioptimus/H-optimus-0"},{"label":"Bioptimus releases (GitHub) \u2014 H-optimus-0 evaluation protocol and results tables","url":"https://github.com/bioptimus/releases/tree/main/models/h-optimus/v0"}],"summary":"1.1-billion-parameter ViT-g/14 pathology encoder trained on more than 500,000 H&E slides (hundreds of millions of tiles), released under Apache-2.0 \u2014 one of the few large pathology foundation models with a permissive licence.","tasks":["feature-extraction","classification","prognosis"],"training":{"institutions":"Bioptimus partner sites (undisclosed list)","size":">500,000 WSIs","summary_en":"Over 500,000 H&E whole-slide images from a multi-site, multi-country collection assembled by Bioptimus; hundreds of millions of tiles.","summary_pl":"Ponad 500 000 preparat\u00f3w H&E z wieloo\u015brodkowej, wielokrajowej kolekcji Bioptimus; setki milion\u00f3w kafelk\u00f3w."},"updated_at":"2026-09-09T21:33:09.703724","url":"/ai-oncology/models/h-optimus-0","usage":{"library":"timm","notes_en":"Gated download (accept terms); the model card specifies the exact normalisation constants \u2014 use them, not ImageNet defaults."},"verified_at":"2026-09-07T17:57:59.478713","verified_by":"Occe3C","version":null,"what_it_does":"Tile embeddings for classification, mutation and biomarker prediction; the Apache licence makes it usable inside commercial pipelines."}