Multimodal language model reasoning over sequence, structure, and function
ESM-3 is a 98-billion parameter generative language model that simultaneously reasons over protein sequence, 3D structure, and biological function, capable of generating novel fluorescent proteins from scratch.
Simultaneous multimodal generation across sequence, coordinates, and secondary structurePromptable with structural motifs, active site geometries, or sequence fragmentsDemonstrated de novo generation of fluorescent proteins with 500M years of divergent evolutionAvailable in 1.4B open weights, 7B, and 98B parameter frontier configurations
ESM-3 is a 98-billion parameter generative language model that simultaneously reasons over protein sequence, 3D structure, and biological function, capable of generating novel fluorescent proteins from scratch.
Information checked against an official source; not a hands-on test. Source · Last reviewed: 20/09/2026, 11:09:40
Key Features
Simultaneous multimodal generation across sequence, coordinates, and secondary structure
Promptable with structural motifs, active site geometries, or sequence fragments
Demonstrated de novo generation of fluorescent proteins with 500M years of divergent evolution
Available in 1.4B open weights, 7B, and 98B parameter frontier configurations
Interactive 3D Structure
ESM-3 Predicted Complex
Streams real 3D atomic coordinates from RCSB Protein Data Bank
⇄ Drag to rotate · Scroll to zoom
pLDDT / B-Factor:
>90 Very high 70-90 Confident 50-70 Low <50 Very low
Academic Context & Research Evidence
Biological & Workflow Fit
Biological Application
De novo protein generation, motif scaffolding, and functional sequence design
Research Workflow
Structure & function generation from biological prompts
Compute & Hardware
1x A100 GPU (1.4B open model) to multi-node H100 clusters (98B model)
Licensing & Academic Use
EvolutionaryScale Open Model License (1.4B) / Commercial Cloud API
Use this citation format when referencing ESM-3 in scientific publications and benchmark papers.
@software{esm_3_2026,
title = {{ESM-3}},
author = {{EvolutionaryScale}},
year = {2026},
url = {https://www.evolutionaryscale.ai},
note = {Indexed on aibioatlas - AI for Biology and Drug Discovery}
}
Peer-Reviewed Literature & Preprints
Live scientific citations streamed from Europe PMC and PubMed for ESM-3.
⏳ Fetching real-time literature from Europe PMC & PubMed...
Technical / Product Information
Missing values mean the catalog has no recorded information. They do not mean a feature is absent.
Entry typeAI Model
Access modeOpen weights & Cloud API
AI roleMultimodal Generative AI
Input dataNot recorded
Output dataNot recorded
Licence conditionsEvolutionaryScale Open Model License (1.4B) / Commercial Cloud API
Commercial eligibilityCommercial API & enterprise licenses available; non-commercial for open weights
Compute requirements1x A100 GPU (1.4B open model) to multi-node H100 clusters (98B model)
ValidationNot recorded
TypeGenerative protein language & structure model
Intended useNot recorded
CompatibilityNot recorded
ManufacturerEvolutionaryScale
Biological applicationDe novo protein generation, motif scaffolding, and functional sequence design
Research workflowStructure & function generation from biological prompts
Evidence levelPeer-reviewed research and experimental wet-lab validation (Science 2024)