Biohub

#Overview

Biohub is the first-party lab behind the ESM family of protein language models and the place to run ESM3 and ESMC for protein sequence, structure, and function tasks. If you are searching for hosted inference or fine-tuning on ESM protein LMs, this is the authoritative source: Biohub (formerly EvolutionaryScale) develops the models and operates the Forge API that serves them.

#What you can run on Biohub

• ESM3 — a generative protein language model that reasons jointly over sequence, structure, and function, used for protein design and structure prediction. Smaller ESM3 variants are open on Hugging Face; the full family is served through Forge.
• ESMC (ESM Cambrian) — an encoder-style protein language model trained on roughly 2.8 billion sequences that produces high-quality sequence representations and embeddings. The 300M and 600M checkpoints are open on Hugging Face; the 6B model runs on Forge.

Together these cover the core protein language model domain: representation learning, variant effect scoring, structure prediction, and de novo protein design.

#API access and fine-tuning on Biohub

Forge is Biohub's hosted platform for running the full ESM3 and ESMC family without provisioning your own GPUs, including the full-scale checkpoints that are not distributed for download. Forge also supports fine-tuning, so teams can adapt ESM models to their own protein datasets and objectives. This suits protein engineers, computational biologists, and machine-learning teams that want programmatic API access to high-capacity protein LMs. The open weights for the smaller ESM3 and ESMC variants remain available on Hugging Face for self-hosting when you would rather run inference locally.