The brief

Analyst view

• ESM3 reasons over protein sequence, structure and function in a generative model trained on 2.78 billion natural proteins.
• Designed esmGFP, a novel fluorescent protein compressing ~500M years of natural evolution into computation.
• Acquired by CZI in Nov 2025 to form Biohub, focusing on curing all diseases by 2100 through AI and biology.

Technical approach

As reported

Multimodal reasoning

Sequence, structure and function together

ESM3 reasons over all three protein properties simultaneously, so scientists can provide mixed constraints and the model completes the protein.

Scale

Massive models on massive compute

Trained on 2.78 billion proteins with 10^24 flops, giving detailed knowledge of how mutations, structures and functions relate.

Open access

API and open-weight versions

ESM3-open available under MIT; AWS and NVIDIA offer secure fine-tuning for enterprise users.