Building towards open medical superintelligence for clinicians
Multimodal foundation models, clinical workflows, and robust evaluations for transparent medical AI.
Spatial biology from histology
Holon-S1
From tissue to gene expression.
Predict spatial gene expression from routine H&E pathology. Explore gene activity in its tissue context and discover interpretable biomarkers.
Building a universal medical foundation model.
Working toward a unified, multimodal understanding of patient state.
Visits, history, symptoms
Blood, urine, chemistry
X-ray, CT, MRI
H&E, IHC, slides
ECG, wearables
DNA, RNA, proteomics
Treatments, response
Lifestyle, context
Model disease trajectory
Predict how a patient's state evolves over time and their risk of adverse outcomes.
Predict intervention response
Identify interventions that can move a patient toward a healthier state.
Accelerate biomedical research
Generate clinically grounded hypotheses from multiscale, longitudinal patient representations.
Patient data
- Clinical Notes
- Labs
- Imaging
- Pathology
- Signals
- Genomics
- Medications
- Demographics
Universal medical AI
Clinical applications
- Disease trajectory
- Intervention response
- Biomedical research
Pushing the frontier with open models and rigorous benchmarking
Explore our available models, benchmark leaderboards, and active open challenges.
Frameworks
Open code for training and evaluating models quickly.
- nanopathTrain and evaluate a pathology foundation model in about an hour on one GPUBlog post
- nanobrainThe nanopath approach, applied to structural brain MRIComing soon
Benchmarks
Open suites for evaluating medical AI.
Foundation models
General-purpose models that other models build on.
Open science
community.
MedARC is the world's largest open medical AI research community — founded by Sophont's CEO and home to thousands of researchers, clinicians, and engineers.
MedARC
The world's largest open medical AI research community.
Join the next discussion.
We invite researchers to present their work, then open the room for discussion with the community. Add the series to your calendar, or sign up below for email notifications.
Publications.
Selected peer-reviewed papers, workshop contributions, and technical reports from the Sophont and MedARC research collective.
Holon-S1 Technical Report
Compress-Distill: Reasoning Trace Compression for Efficient Knowledge Distillation
Medmarks: A comprehensive open-source LLM benchmark suite for medical tasks
Scaling Vision Transformers for Functional MRI with Flat Maps
MIRAGE: Robust multi-modal architectures translate fMRI-to-image models from vision to mental imagery
OpenMidnight: How to Train a State-of-the-Art Pathology Foundation Model with $1.6k
NSD-Imagery: A benchmark dataset for fMRI mental imagery
MindEye2: Shared-Subject Models enable fMRI-to-Image with 1 hour of data
A vision–language foundation model for realistic chest X-ray generation
CheXAgent: A vision-language foundation model to enhance efficiency of chest X-ray interpretation
Reconstructing the mind’s eye: fMRI-to-image with contrastive learning
Built by researchers from the field.
A small team of medical AI researchers with roots in academic hospitals, open-source ML, and NeuroAI.
Tanishq Mathew Abraham, Ph.D.
CEO & Co-Founder
Former Research Director at Stability AI; founded MedARC, the world’s largest online medical AI research community. Graduated at 19 with a Ph.D. in Biomedical Engineering at University of California, Davis.
- UC Davis
- Stability AI
Paul Scotti, Ph.D.
CTO & Co-Founder
Former Head of NeuroAI at Stability AI and postdoc at Princeton University. Paul brings over a decade of experience in computational neuroscience and was first author of the MindEye and MindEye2 papers on reconstructing perception from brain activity.
- Princeton
- Stability AI
Hassan Muhammad, Ph.D.
Head of Applied Research
Previously Senior Director of AI at Pathomiq and Research Scientist at Memorial Sloan Kettering Cancer Center. Ph.D. from Weill Cornell Medicine
- Memorial Sloan Kettering
- Pathomiq
- Cornell
Anthony Cessna
Head of Commercial Strategy and Data Partnerships
Previously led Data Business Development at Foundation Medicine and was a Key Account Director at Tempus AI. Anthony brings experience partnering with biopharma to advance precision medicine using real-world data and AI.
- Tempus
- Foundation Medicine
- Fred Hutchinson
Benjamin Warner
Founding Research Scientist
A self-taught AI researcher, Benjamin was the project lead and first author of ModernBERT at Answer.AI. At Sophont, he leads medical LLM and agent research.
- Answer.ai
- Duke
Connor Lane
Founding Research Scientist
Former neuroscientist at the Child Mind Institute, Connor leads Sophont’s neuroimaging foundation model research.
- Child Mind Institute
- Johns Hopkins
The listed profiles are joined by a broader founding group of scientists and engineers spanning LLMs, NeuroAI, and medical AI. See open roles
In the press.
Selected coverage and appearances.
OpenAI mentions Medmarks as part of their ChatGPT for Clinicians announcement
Sophont announces $9.22M seed round
NY Times discusses our MindEye research
[Mind]ography: the world's first printed memories
The Mind-Reading Revolution with Tanishq
Open Foundation Models for the Future of Medicine
US Senate hearing references our medical AI work
Our investors.
We are fortunate to partner with investors and strategic allies across medicine, AI research, and venture.
Delphi Ventures
Brightwing Capital
Jeff Dean
Google Chief Scientist
Logan Kilpatrick
Google DeepMind
Clem Delangue
Hugging Face CEO via Factorial Capital
Lukas Biewald
Weights & Biases co-founder
Mei Zuo
Mei Ventures
Let's build the open future of medicine.
We partner with healthcare providers, biopharma and life sciences companies, research institutions, and mission-aligned investors.