ESM3 is a groundbreaking generative AI model for biology developed by EvolutionaryScale.
It's a language model capable of generating novel proteins, trained on a massive dataset of 2.78 billion proteins across Earth's natural diversity.
ESM3 was trained using 1 trillion teraflops of compute power, more than any other known model in biology.
It's the first generative model for biology that simultaneously reasons over the sequence, structure, and function of proteins, enabling scientists to understand and create new proteins.
The model demonstrated its capabilities by generating a new Green Fluorescent Protein (GFP), simulating 500 million years of evolution.
ESM3 contains 98 billion parameters, representing a significant advancement over its predecessor, ESM2.
It has potential applications in drug discovery, materials science, carbon capture, and other areas of biological research.
EvolutionaryScale is making ESM3 available through an API for closed beta access, and a smaller open version for non-commercial use.
The model will be integrated with Amazon Web Services (AWS) and NVIDIA platforms to accelerate its applications in various fields.
ESM3 represents a significant step towards making biology programmable and has the potential to accelerate scientific discovery across a broad range of applications in life sciences.