BIOLOGY.DIGITAL
EvolutionaryScale Releases ESM3 98B Generative Biology Model for Programmable Protein Design

Image: Science

Protein Design

EvolutionaryScale Releases ESM3 98B Generative Biology Model for Programmable Protein Design

July 8, 2026 • Source: Science

EvolutionaryScale unveiled ESM3, a 98-billion-parameter frontier multimodal generative model trained across sequence, 3D atomic structure, and protein function tokens across 2.78 billion natural proteins.

**Executive Summary (2026-07-08):** EvolutionaryScale unveiled ESM3, a 98-billion-parameter frontier multimodal generative model trained across sequence, 3D atomic structure, and protein function tokens across 2.78 billion natural proteins. Source: Science.

Technical Architecture & R&D Impact

ESM3 simultaneously reasons over sequence, structure, and function in a unified discrete token space. In peer-reviewed benchmarks published in Science, the model generated esmGFP, a functional fluorescent protein sharing only 58% sequence identity with its closest natural homolog, equivalent to simulating over 500 million years of natural evolution in silico. Enterprise biopharma teams can deploy open-weights ESM3-open via NVIDIA BioNeMo and AWS HealthOmics.

Enterprise Biopharma Procurement Takeaways

For computational biology leads, medicinal chemists, and R&D IT architects, this milestone underscores the rapid convergence of wet-lab experimental validation and scalable cloud foundation models during H2 2026. Teams evaluating vendors in the Protein Design category should benchmark API reproducibility, GxP audit readiness, and zero-data-retention VPC deployment terms.

Related Products

Published July 8, 2026

More News

Last updated: July 8, 2026

Ask AI