EvolutionaryScale

EvolutionaryScale is a frontier AI company focused on the life sciences, building powerful generative and representation models that enable scientists to understand, imagine, and create proteins. The company develops artificial intelligence that deepens biology's understanding at a molecular level, with the goal of benefiting human health and society through open, safe, and responsible research and scientific community partnerships.

At the core of EvolutionaryScale's platform is ESM3, a multimodal generative model that simultaneously reasons over the fundamental properties of a protein — sequence, structure, and function. Biology is treated as a programmable system: every living organism shares the same genetic code across the same 20 amino acids, and ESM3 is trained to understand, translate, and generate across this biological language. Users can input any combination of sequence, structure, or function data to prompt ESM3 to explore a vast space of protein possibilities.

ESM3 — Generative Protein Model

• Reasons simultaneously over protein sequence, structure, and function, allowing flexible multimodal prompting.
• Capable of designing entirely novel proteins beyond what exists in nature — demonstrated through the creation of esmGFP, a novel Green Fluorescent Protein generated via chain-of-thought prompting.
• esmGFP represents a vast evolutionary departure from natural fluorescent proteins, estimated to be equivalent to 500 million years of natural evolution.
• Available as a family of models in three sizes: small, medium, and large.
• Trained on 771 billion unique tokens of training data, drawing from 2.78 billion natural proteins sampled from diverse organisms and biomes.
• The largest model contains 98 billion parameters and was trained using 1 trillion teraflops (10^24 FLOPs) of computing power.

ESM Cambrian — Representation Learning Model

• A parallel model family to ESM3, focused on protein sequence representation learning rather than generation.
• Delivers breakthrough performance and efficiency, setting a new state of the art for representation learning in protein science.
• Designed to complement the generative capabilities of ESM3 with best-in-class sequence understanding.

Deployment and Access Options

• Forge API: All ESM3 models are available through EvolutionaryScale's Forge platform, currently in closed Beta with access available by application.
• AWS: ESM3 is available on AWS via SageMaker, with planned availability on the AWS Omics platform and AWS Bedrock.
• NVIDIA: ESM3 will be available on BioNemo and as an NVIDIA NIM microservice, optimized in collaboration with NVIDIA.
• Open Model: ESM3-open is a small but powerful model with weights and source code available on GitHub under a non-commercial license, making it accessible to the broader scientific community.

Responsible Development and Mission

• All models are built and deployed in accordance with EvolutionaryScale's responsible development framework.
• The company is committed to scientific rigor, responsible AI development, and advancing the frontiers of scientific discovery.
• EvolutionaryScale pursues open and safe research, with a focus on scientific community partnerships.
• The company is joining forces with Biohub, further expanding its collaborative scientific mission.

EvolutionaryScale represents a new generation of AI-driven tools for protein science, combining massive scale training, generative modeling, and responsible deployment to empower researchers working at the frontier of biology and medicine.