HCLS Foundation Models Map
I track foundation models across health and life sciences and judge every one against the same rubric, but the calls are mine. Treat it as a well-kept notebook, not gospel. Last updated Nov 2026.
178foundation models tracked
12categories in 4 areas
81%have downloadable weights
Counts every model not tagged private, meaning closed, commercial or gated. If a gated model is missing that tag, this number runs high.
How fast the field grew
Each model line counts from the year its first version came out, even though the list below shows only the latest versions.
So Evo 2 (2025) counts from 2024, when the first Evo came out.
BiologyChemistryClinicalEmerging
Who builds them
The 12 organisations with the most models. A lab counts toward its university, and DeepMind toward Google.
Microsoft12
Stanford12
Harvard11
NVIDIA11
Tsinghua10
Google8
Shanghai Jiao Tong6
Tencent5
Arc Institute4
CUHK4
U Toronto4
MIT3
Biology 25
Chemistry 4
Clinical 31
All 178 models
| Model | Category | Modality | Description | Tags |
|---|---|---|---|---|
| Biology | ||||
| AlphaGenome | Genomics | DNA | DeepMind regulatory-genomics model for variant-effect prediction across modalities including expression, splicing, chromatin, and contact maps. | |
| Borzoi | Genomics | DNA | Convolution-transformer predicting RNA-seq coverage from DNA sequence. | |
| Carbon-X | Genomics | DNA | Open 14B language model for DNA understanding and generation, from the Hugging Face Carbon family. | |
| DNABERT-2 | Genomics | DNA | Multispecies genome language model using byte-pair-encoding tokenization. | |
| DNAGPT | Genomics | DNA | Generalized pre-trained tool for multiple DNA sequence analysis tasks. | |
| Evo 2 | Genomics | DNA | 7B and 40B-parameter DNA model that processes up to 1M base pairs at nucleotide resolution; backbone of the Mayo Clinic EVEE variant pathogenicity work | |
| gLM | Genomics | DNA | Genomic language model that predicts protein co-regulation and function from contig context. | |
| GPN-MSA | Genomics | DNA | Alignment-based DNA language model for genome-wide variant effect prediction. | |
| HyenaDNA | Genomics | DNA | Long-context DNA sequence model based on the Hyena architecture | |
| ModernGENA | Genomics | DNA | ModernBERT-based DNA foundation model (up to 377M) positioned as an efficient encoder baseline. | |
| Nucleotide Transformer v3 | Genomics | DNA | Multispecies DNA foundation model family (8M to 650M) with long-context pretraining, post-trained and generative variants. | |
| 3UTRBERT | Transcriptomics | RNA | Language model over mRNA 3' UTR sequences for post-transcriptional regulation tasks. | |
| ATOM-1 | Transcriptomics | RNA | RNA foundation model trained on chemical mapping data for structure and function. | |
| CellPLM | Transcriptomics | scRNA | Cell language model pre-trained beyond single cells, using spatially resolved transcriptomics. | |
| CodonBERT | Transcriptomics | RNA | Language model over codon sequences for mRNA property prediction. | |
| ERNIE-RNA | Transcriptomics | RNA | RNA language model with structure-enhanced representations. | |
| GeneCompass | Transcriptomics | scRNA | Knowledge-informed cross-species single-cell foundation model for gene regulation. | |
| Geneformer | Transcriptomics | RNA | Transformer pre-trained on about 104M single-cell human transcriptomes (V2); widely used for drug-target and cell-state inference. | |
| GenePT | Transcriptomics | scRNA | Single-cell and gene embeddings derived from gene-description text via an LLM. | |
| GenerRNA | Transcriptomics | RNA | Generative pre-trained language model for de novo RNA design. | |
| Orthrus | Transcriptomics | RNA | Evolutionary and functional RNA foundation model. | |
| RiNALMo | Transcriptomics | RNA | General-purpose RNA language model that generalizes to structure prediction tasks. | |
| RNA-FM | Transcriptomics | RNA | Interpretable RNA foundation model trained on unannotated ncRNA for structure and function prediction. | |
| RNA-MSM | Transcriptomics | RNA | MSA-based RNA language model leveraging homologous sequences for structure signal. | |
| RNABERT | Transcriptomics | RNA | Informative RNA base embedding via masked LM for structural alignment and clustering. | |
| scBERT | Transcriptomics | RNA | BERT-style single-cell RNA-seq foundation model for cell type annotation | |
| scFoundation | Transcriptomics | scRNA | Large-scale foundation model on single-cell transcriptomics. | |
| scGPT | Transcriptomics | RNA | Single-cell foundation model encoding gene and cell relationships across large atlases | |
| SCimilarity | Transcriptomics | scRNA | Cell-embedding foundation model for rapid similar-cell-profile retrieval. | |
| scMulan | Transcriptomics | scRNA | Multitask generative pre-trained language model for single-cell analysis. | |
| scPRINT | Transcriptomics | scRNA | Single-cell RNA FM pre-trained on 50M cells for robust gene network prediction. | |
| SpliceBERT | Transcriptomics | RNA | Language model pretrained on precursor-mRNA for RNA splicing prediction. | |
| Stack | Transcriptomics | RNA | Arc Institute virtual-cell effort focused on in-context learning over cellular conditions. | |
| State | Transcriptomics | RNA | Arc Institute virtual-cell model aimed at perturbation prediction over cellular conditions. | |
| TranscriptFormer | Transcriptomics | RNA | Cross-species generative single-cell transcriptomics model from the Chan Zuckerberg Initiative. | |
| UNI-RNA | Transcriptomics | RNA | Universal pre-trained RNA representation model. | |
| Universal Cell Embeddings | Transcriptomics | scRNA | Single-cell foundation model producing universal cell representations across species. | |
| UTR-LM | Transcriptomics | mRNA -UTR | 5' UTR language model for decoding untranslated regions of mRNA. | |
| xTrimoGene | Transcriptomics | scRNA | Efficient and scalable representation learner for single-cell RNA-seq data. | |
| AlphaFold 3 | Protein | protein | Atomic structures of proteins DNA RNA and ligand complexes | |
| Ankh | Protein | protein | Optimized protein language model for general-purpose modelling. | |
| BioEmu-2 | Protein | protein -conformation | Diffusion model (~500M) that generates all-atom conformational ensembles for proteins, complexes, ligands and nucleic acids. | |
| Boltz-2 | Protein | protein | Open-source biomolecular co-folding model that also predicts small-molecule binding affinity. | |
| CaLM | Protein | codon | Codon language embeddings for protein engineering. | |
| CARP | Protein | protein | Convolutional protein sequence model competitive with transformers. | |
| Chai-1 | Protein | protein | Open biomolecular structure prediction model for complexes. | |
| ESM-3 | Protein | protein | Multimodal protein language model over sequence structure and function | |
| ESMFold2 | Protein | protein -structure | Single-sequence protein structure predictor built on ESM protein language model embeddings. | |
| EvoDiff | Protein | protein | Sequence-based diffusion generative model for protein design. | |
| Evolla | Protein | protein + text | Decodes molecular language of proteins for question answering and annotation. | |
| GearNet | Protein | protein -structure | Geometric structure pretraining for protein representation learning. | |
| HelixFold-Single | Protein | protein -structure | MSA-free protein structure prediction using a protein language model. | |
| MSA Transformer | Protein | protein | Protein language model with axial attention over multiple-sequence alignments. | |
| OmegaFold | Protein | protein | Predicts high-resolution protein structure from a single sequence without MSAs. | |
| OntoProtein | Protein | protein | Protein pretraining with gene ontology embeddings. | |
| OpenFold3 | Protein | protein | Open-source effort to support reproducible biomolecular co-folding models. | |
| PrimateAI-3D | Protein | protein | Variant-effect prediction model integrating 3D structure | |
| Profluent E1 | Protein | protein | Retrieval-augmented protein language model family (150M to 600M) that conditions on homologous sequences. | |
| ProGen2 | Protein | protein | Family of autoregressive protein language models up to 6.4B parameters for design and fitness. | |
| ProteinBERT | Protein | protein | Universal deep-learning model of protein sequence and function with GO annotation pretraining. | |
| ProteinDT | Protein | protein | Text-guided protein design coupling sequence and language representations. | |
| ProtGPT2 | Protein | protein | Deep unsupervised generative language model for protein design. | |
| ProTrek | Protein | protein + structure + text | Tri-modal contrastive learning across protein sequence, structure, and text. | |
| ProtST | Protein | protein + text | Multi-modality learning of protein sequences and biomedical texts. | |
| ProtTrans | Protein | protein | Family of protein language models (T5, BERT, Albert, XLNet, Electra) trained via self-supervision at HPC scale. | |
| RFD4-Proteína | Protein | protein | Flow-matching model that co-designs protein sequence and all-atom structure conditioned on targets, motifs or ligands. | |
| RoseTTAFold All-Atom | Protein | protein | Open structure-prediction and design model from the Baker Lab, extending RoseTTAFold to broader biomolecular assemblies including proteins, nucleic acids, small molecules, and covalent modifications. | |
| SaProt | Protein | protein + structure | Protein language model with structure-aware vocabulary. | |
| Uni-Fold | Protein | protein | Open reimplementation and training of AlphaFold-style protein structure prediction. | |
| UniRep | Protein | protein | Unified rational protein engineering with sequence-based deep representation learning. | |
| xTrimoPGLM | Protein | protein | Unified 100B-scale pre-trained transformer for protein language. | |
| EpiGePT | Epigenomics | epigenome | Pretrained transformer for epigenomics. | |
| Pleiades | Epigenomics | cfDNA | Epigenetic foundation model for cell-free DNA currently focused on Alzheimer's biomarker discovery | |
| AbLang2 | Antibody and biologics | antibody -sequence | Antibody-specific language model for infilling and humanization | |
| AntiBERTa2 | Antibody and biologics | antibody -sequence | Antibody-specific transformer language model | |
| AntiBERTy | Antibody and biologics | antibody -sequence | BERT-style antibody language model trained on the Observed Antibody Space | |
| IgLM | Antibody and biologics | antibody -sequence | Generative antibody language model for heavy and light chain design | |
| Cell2Sentence | Multi-omics and systems | scRNA + text | Teaches LLMs the language of biology by rendering scRNA cells as sentences. | |
| KRONOS | Multi-omics and systems | spatial -proteomics | Foundation model for spatial proteomics, learning cell and tissue representations from multiplex protein imaging. | |
| Nicheformer | Multi-omics and systems | scRNA + spatial | Single-cell and spatial omics foundation model for tissue context modeling. | |
| Chemistry | ||||
| ChemBERTa-2 | Small-molecule | SMILES | Transformer pretrained on ~77M SMILES for molecular property prediction. | |
| ChemGPT | Small-molecule | SMILES | Large-scale generative model for chemistry over SMILES | |
| GROVER (molecule) | Small-molecule | molecular -graph | Self-supervised graph-transformer pretraining on large-scale molecular data. | |
| MegaMolBART | Small-molecule | SMILES | BART-style model for SMILES part of the BioNeMo umbrella | |
| MoleculeSTM | Small-molecule | SMILES | Multimodal molecule-text model linking chemical structures and natural language. | |
| MoLFormer | Small-molecule | SMILES | Large-scale chemical language model over SMILES originally explored with AstraZeneca | |
| MolMIM | Small-molecule | SMILES | Controlled molecule generation with property guidance | |
| Clinical | ||||
| BiomedCLIP | Medical imaging | vision -language | Biomedical vision-language model pretrained on 15M PubMed image-text pairs | |
| BioViL | Medical imaging | vision -language | Joint chest X-ray image and radiology-report representation learning. | |
| CheXFound | Medical imaging | radiograph | Chest X-ray foundation model | |
| CheXzero | Medical imaging | radiograph | Self-supervised chest X-ray model for zero-shot pathology detection from reports. | |
| Clinical-BERT (VLM) | Medical imaging | vision -language | Vision-language pretraining for radiograph diagnosis and report generation. | |
| CLIP-Driven Universal Model | Medical imaging | CT | CLIP-guided universal model for multi-organ CT segmentation and tumor detection. | |
| ConVIRT | Medical imaging | vision -language | Contrastive image-text pretraining of medical images from paired radiology reports. | |
| CT-CLIP | Medical imaging | CT | Contrastive foundation model on paired 3D chest CT volumes and radiology reports. | |
| CXR Foundation | Medical imaging | radiograph | Chest X-ray foundation model from Google | |
| GLoRIA | Medical imaging | vision -language | Global-local contrastive image-text pretraining for chest X-ray. | |
| KAD | Medical imaging | radiograph | Knowledge-augmented chest X-ray diagnosis via image-text pretraining. | |
| KoBo | Medical imaging | vision -language | Knowledge-boosted contrastive vision-language pretraining for radiology. | |
| LLaVA-Med | Medical imaging | vision -language | Biomedical vision-language assistant adapted from LLaVA on PMC figure-caption data. | |
| LVM-Med | Medical imaging | medical -imaging | Self-supervised large vision model pretrained on 1.3M medical images. | |
| MAIRA-2 | Medical imaging | vision -language | Grounded radiology report generation from chest X-rays. | |
| Med-Flamingo | Medical imaging | vision -language | Few-shot multimodal medical model adapted from Flamingo. | |
| Med-Gemini | Medical imaging | multimodal | Multimodal medical foundation model for radiology report generation and visual question answering | |
| Med3D | Medical imaging | 3D -medical -imaging | Pretrained 3D CNN backbones for volumetric medical imaging (MedicalNet). | |
| MedCLIP | Medical imaging | vision -language | Contrastive vision-language pretraining from decoupled medical images and reports. | |
| MedKLIP | Medical imaging | vision -language | Knowledge-enhanced medical image-text pretraining for radiology. | |
| MedSAM2 | Medical imaging | medical -image -segmentation | Promptable segmentation foundation model for 3D medical images and videos. | |
| MedVersa | Medical imaging | vision -language | Generalist learner for multifaceted medical image interpretation and reporting. | |
| Models Genesis | Medical imaging | 3D -medical -imaging | Self-supervised pretrained 3D models for medical image analysis. | |
| NV-Reason-CT | Medical imaging | CT -3D | 3D CT vision-language model for radiology reports, question answering and multi-step reasoning on chest and abdominal CT. | |
| PMC-CLIP | Medical imaging | vision -language | CLIP pretrained on PubMed Central figure-caption biomedical pairs. | |
| PubMedCLIP | Medical imaging | vision -language | CLIP fine-tuned on PubMed image-text pairs for medical visual question answering. | |
| RAD-DINO | Medical imaging | radiograph | Image-only DINOv2 chest X-ray encoder trained without text supervision. | |
| RadFM | Medical imaging | radiology | Radiology foundation model for general medical imaging tasks | |
| RETFound | Medical imaging | retina | MAE-pretrained retinal-image foundation model for ocular and systemic disease. | |
| SAM-Med3D | Medical imaging | 3D -medical -imaging | Fully-learnable 3D SAM-style foundation model for volumetric segmentation. | |
| SegVol | Medical imaging | CT | Promptable volumetric CT segmentation foundation model. | |
| STU-Net | Medical imaging | CT | Scalable nnU-Net pretrained on TotalSegmentator for transferable CT segmentation. | |
| UniverSeg | Medical imaging | medical -image -segmentation | Universal in-context medical image segmentation across unseen tasks. | |
| USFM | Medical imaging | ultrasound | Spatial-frequency dual-masked self-supervised ultrasound foundation model. | |
| VisionFM | Medical imaging | ophthalmic -imaging | Multi-task ophthalmology foundation model across eye-imaging modalities. | |
| VISTA3D | Medical imaging | CT | Unified 3D CT foundation model for automatic and interactive segmentation. | |
| CHIEF | Digital pathology | WSI | Generalizable whole-slide pathology foundation model. | |
| CONCH | Digital pathology | vision -language | Vision-language pathology model from the Mahmood Lab | |
| CTransPath | Digital pathology | WSI | Self-supervised transformer backbone for histopathology (TransPath line). | |
| H-optimus-1 | Digital pathology | WSI | Bioptimus histology foundation model (1.1B-parameter ViT) for downstream pathology tasks. | |
| MUSK | Digital pathology | vision -language | Vision-language pathology foundation model for precision oncology. | |
| PathChat | Digital pathology | vision -language | Multimodal conversational assistant for whole-slide pathology. Weights gated. | |
| PLIP | Digital pathology | WSI | Pathology vision-language CLIP trained on OpenPath social-media histopathology. | |
| PRISM2 | Digital pathology | WSI | Paige's second-generation multimodal slide-level pathology model, trained with clinical dialogue. | |
| Prov-GigaPath | Digital pathology | WSI | Gigapixel pathology foundation model with tile-to-slide hierarchical representations | |
| QuiltNet | Digital pathology | vision -language | Pathology CLIP trained on Quilt-1M video-derived image-text pairs. | |
| RudolfV | Digital pathology | WSI | DINOv2 pathology foundation model built with pathologist guidance. Weights closed. | |
| TITAN | Digital pathology | vision -language | Multimodal whole-slide pathology foundation model, multi-site validated. | |
| UNI2-h | Digital pathology | WSI | General-purpose pathology foundation model (ViT-H) from the Mahmood Lab, trained on over 200M tiles from H&E and IHC slides. | |
| Virchow2 | Digital pathology | WSI | Paige's second-generation pathology foundation model, scaled with mixed-magnification self-supervised training. | |
| AMIE | Clinical language and patient | medical -text | Google Research diagnostic-dialogue research system for medical reasoning and conversations. | |
| BEHRT | Clinical language and patient | EHR | Transformer over structured EHR diagnosis-code sequences to predict future diagnoses. | |
| Bio_ClinicalBERT | Clinical language and patient | clinical -text | Clinical BERT pretrained on MIMIC-III notes, initialized from BioBERT. | |
| BioBART | Clinical language and patient | biomedical -text | BART sequence-to-sequence model pretrained on biomedical literature for generative NLP. | |
| BioBERT | Clinical language and patient | biomedical -text | BERT pretrained on PubMed and PMC for biomedical text mining. | |
| BioGPT | Clinical language and patient | biomedical -text | Generative biomedical transformer pretrained on PubMed literature for text generation and mining. | |
| BioLinkBERT | Clinical language and patient | biomedical -text | Document-link-aware biomedical language model pretrained on PubMed. | |
| BiomedGPT | Clinical language and patient | vision -language | Generalist biomedical transformer spanning vision, language, and multimodal tasks. | |
| BioMedLM | Clinical language and patient | biomedical -text | 2.7B GPT trained on PubMed for biomedical language (also called PubMedGPT). | |
| BioMegatron | Clinical language and patient | biomedical -text | Large Megatron-style biomedical language model pretrained on PubMed-scale corpora. | |
| BioMistral | Clinical language and patient | biomedical -text | Mistral-7B further pretrained on PubMed Central for biomedical domains. | |
| ChatDoctor | Clinical language and patient | medical -text | LLaMA finetuned on patient-doctor medical dialogue. | |
| ClinicalBERT (Huang) | Clinical language and patient | EHR -text | BERT modeling clinical notes for hospital-readmission prediction. | |
| ClinicalGPT | Clinical language and patient | clinical -text | LLM fine-tuned on clinical records and medical dialogue with multi-task evaluation. | |
| ClinicalT5 | Clinical language and patient | clinical -text | T5 encoder-decoder pretrained on MIMIC clinical text. Weights gated. | |
| CLMBR | Clinical language and patient | EHR -codes | Clinical Language Model Based Representations from Stanford's Shah lab | |
| GatorTron | Clinical language and patient | EHR -text | Clinical LM backbone for downstream NLP at UF Health | |
| GatorTronGPT | Clinical language and patient | EHR -text | Generative clinical LLM trained on de-identified EHR notes; generative sibling of GatorTron. | |
| HuatuoGPT-II | Clinical language and patient | clinical -text | Baichuan-2-based Chinese medical LLM for consultation and question answering. | |
| KeBioLM | Clinical language and patient | biomedical -text | Knowledge-enhanced biomedical language model integrating biomedical entity knowledge. | |
| Me-LLaMA | Clinical language and patient | biomedical -text | LLaMA-2 medical foundation family for clinical NLP and reasoning tasks. | |
| Med-BERT | Clinical language and patient | EHR | BERT pretrained on structured EHR diagnosis-code sequences for clinical prediction. | |
| Med-PaLM 2 | Clinical language and patient | medical -text | PaLM-2 model tuned for medical question answering; successor to Med-PaLM. Weights closed. | |
| MedAlpaca | Clinical language and patient | medical -text | Open medical-QA LLM collection finetuned on LLaMA. | |
| MedCPT | Clinical language and patient | biomedical -text | Contrastively pretrained biomedical text encoder for PubMed retrieval. | |
| MedGemma 1.5 | Clinical language and patient | vision -language | Google open medical vision-language model (4B) for imaging and clinical text tasks. | |
| Meditron3-70B | Clinical language and patient | medical -text | EPFL's third-generation 70B medical language model for clinical reasoning. | |
| NYUTron | Clinical language and patient | EHR -text | Clinical language model trained on NYU Langone clinical notes | |
| PMC-LLaMA | Clinical language and patient | biomedical -text | LLaMA further pretrained on PubMed Central papers and medical books. | |
| PubMedBERT | Clinical language and patient | biomedical -text | BERT pretrained from scratch on PubMed abstracts and full text (aka BiomedBERT). | |
| Radiology-Llama2 | Clinical language and patient | radiology -text | Llama-2 model specialized for radiology report generation and understanding. | |
| Standard Model Bio | Clinical language and patient | EHR -multi -omics | Sits between EHR foundation models and multi-omics patient-trajectory modeling, learning unified patient representations for stratification and clinical-development use cases. | |
| Truveta Language Model | Clinical language and patient | EHR -text | Trained on the largest linked EHR corpus in the United States; clinical reasoning over longitudinal records | |
| Endo-FM | Surgical video | video | Endoscopy foundation model with video pre-training for downstream endoscopic tasks | |
| SurgVISTA | Surgical video | video | Surgical video spatio-temporal transformer for phase recognition and instrument tracking | |
| SurgVLP | Surgical video | vision -language | Surgical vision-language pre-training | |
| Emerging | ||||
| BioT5+ | Emerging | molecules + protein + text | Generalized biological understanding with IUPAC integration and multi-task tuning. | |
| HeartBEiT | Emerging | ECG | ECG-specific foundation model exploring transformer pretraining for cardiac waveforms. | |
| LaBraM | Emerging | EEG | Large-scale pretraining for EEG signals; representative early entry in the brain-waveform foundation-model space. | |
| METAGENE-1 | Emerging | DNA -metagenomic | Arc Institute foundation model for metagenomic sequence, oriented toward microbiome and pathogen-surveillance applications. | |
| TxGNN | Emerging | knowledge -graph | Graph foundation model for zero-shot drug repurposing over a biomedical knowledge graph. | |