Open Athena
Open Athena is a nonprofit that brings frontier AI capabilities to academia by openly developing and sharing scientific foundation models. Its PlantCAD model, built with Cornell, interprets plant DNA to identify genes influencing growth and yield, while BoltzGen, developed with MIT, designs proteins that bind to user-specified targets. The Marin initiative conducts all research in the open, sharing infrastructure and training knowledge for decentralized collaboration.
- Artificial Intelligence
- Biotechnology
- AI Agents
- Drug Discovery & Therapeutics
- Healthcare Technology
- Software Only
Funding
Founders
Founder details are not available yet.
Product
Problem
Academic researchers often lack access to the advanced foundation models and training infrastructure available in industry, limiting their ability to apply AI to complex scientific problems. The inner workings of frontier models remain opaque, and open-weight alternatives do not share the data or exact training methodology, hindering reproducibility and collaborative scientific progress.
Solution
Open Athena is a nonprofit that brings frontier AI capabilities to academia by developing and openly sharing specialized scientific foundation models. Its PlantCAD model, built with Cornell, learns to interpret plant DNA to enable identification of genes influencing growth, yield, and resilience, while GeneCAD applies this knowledge for gene discovery. BoltzGen, developed with MIT, is a generative model that designs proteins binding to user-specified targets. Through the Marin initiative, Open Athena conducts all model research and development openly—sharing infrastructure, experimental results, and training knowledge so that anyone in the world can contribute to or leverage the work.
Target Audience
The primary audience is academic research laboratories, computational biologists, and engineers in universities and research institutions who need cutting-edge foundation models and transparent training methodologies to advance work in genomics and protein science.
Features
- PlantCAD: a DNA foundation model that reads plant genomic sequences and is trained in partnership with Cornell University
- GeneCAD: a model component that identifies genes responsible for shaping plant growth, yield, and resilience traits
- BoltzGen: a target-specific protein generation model developed with MIT that designs novel proteins with user-defined binding properties
- Marin: an open research paradigm that conducts every experiment in the open, disclosing all discussions, twists, and turns for full scientific transparency
- Shared infrastructure for running experiments reliably and efficiently, plus the science of how to train models that only exists in frontier labs
- Scalable and decentralized collaboration model enabling anyone in the world to contribute to and leverage the work; early applications already include training DNA models
- MarinDNA: a 1B genomic language model competitive with Evo 2 40B while using ~1,980× fewer training FLOPs and scoring variants ~2,330× faster
- Marin's largest training run: a 535B total parameter, 23B active parameter mixture-of-experts model trained on 18 trillion tokens, with publicly predicted performance benchmarks