EvolutionaryScale

External ProteinGuide researchers' public experimental repository documents using ESM3 together with FMIF to design TadA-derived adenine base editors. It lists unguided and guided experimental libraries with two and three biological replicates, respectively, and released enrichment data and reproduction scripts. This external experimental research use is distinct from EvolutionaryScale's own esmGFP generation experiment.

EvolutionaryScale reports two laboratory experiments of 96 ESM3-generated proteins each producing esmGFP, a green fluorescent protein with brightness similar to natural GFPs and 58% sequence similarity to its closest known fluorescent counterpart.

EvolutionaryScale applies ESM-family biological foundation models to protein design and life-science workflows, while broad commercial deployment remains early.

EvolutionaryScale develops frontier biological foundation models for protein understanding and generation.

EvolutionaryScale develops frontier biological foundation models for protein understanding and generation.

EvolutionaryScale ESM3 is a 98B-parameter multimodal generative model that jointly reasons over protein sequence, structure and function, trained on 2.78B natural proteins and 771B unique tokens with more than 1e24 FLOPs; it generated a novel fluorescent protein representing an estimated 500M years of evolutionary distance.

EvolutionaryScale's official site now states that it is part of Biohub and routes current ESM APIs, recruiting and media contacts to Biohub; the site says current services are governed by Biohub terms and privacy policy.

EvolutionaryScale has assembled specialized frontier-model and computational-biology talent with substantial research capacity.

EvolutionaryScale's ESM Cambrian report describes learning protein representations by predicting amino acids from unlabeled sequences, with 300M, 600M and 6B parameter scales and compute-scaling experiments linking representations to structure and function.

Foundation-model engineering transfers broadly, while computational biology specialization limits demonstrated cross-market portability.

Foundation-model engineering transfers broadly, while computational biology specialization limits demonstrated cross-market portability.