ESM-2
Protein language model for embeddings, fine-tuning and variant scoring.
Developer: Meta AI (FAIR)
Overview
- Developer
- Meta AI (FAIR)
- Category
- Protein language models
- Commercial rights
- Commercial OK
- Commercial cost
- Free commercially
- Compute
- Single GPU
- Open weights
- Yes
- Last verified
Capabilities
Protein language model. Input: a protein sequence; output: per-residue and per-sequence embeddings, masked-token likelihoods and variant pseudo-likelihood scores - the standard feature extractor for fitness/liability models and design filters.
When to use it
Default embedding model for property predictors and sequence ranking; the 650M checkpoint runs on one GPU. Its likelihoods track evolutionary plausibility, not binding affinity or developability.
Input / Output
- Input
- a protein sequence
- Output
- per-residue and per-sequence embeddings, masked-token likelihoods and variant pseudo-likelihood scores - the standard feature extractor for fitness/liability models and design filters
Compute
Single GPU. Other tools with this compute profile →
Commercial Licensing
Commercial rights — may a company use it?
Commercial use OKLicence as writtenMIT License
The Hugging Face model card front-matter declares 'license: mit' for the released weights; the source repo LICENSE is also MIT.
The licence read at the source allows use by a company. Attribution or copyleft conditions may still apply — read the note.
Commercial cost — is commercial use free of charge?
Free for commercial useNothing to pay: the licence read at the source grants commercial use with no fee, and the code and weights are public downloads. You still pay for your own compute.
Price and rights are separate: this panel is about money, the panel above is about permission.
- Commercial OK ≠ Free of charge. A tool can be licensed for company use and still need a paid licence for a dependency (BindCraft → PyRosetta).
- Free download ≠ Commercially licensed. Weights you can download at no cost may still forbid commercial use (AlphaFold 3).
- Unverified ≠ Commercially permitted. If no published statement settles it, treat it as unknown — not as a yes.
- Every component counts. Each status considers the code licence, the model-weights terms, required dependencies and any terms of use.
Model Weights
Open weights. A separate weights licence is Not documented as its own field; where weights carry different terms from the code, that is stated in the licence explanation above.
Dependencies
Not documented as a separate field. Any dependency that affects commercial use (for example a paid PyRosetta licence) is named in the licence and cost explanations above.
Official References
- Official site / repo
- huggingface.co/facebook/esm2_t33_650M_UR50D
From the makers of FoldScout
From AI Design to Wet-Lab Validation
Designed a promising protein or antibody? Seq2Study helps translate AI-generated sequences into experimental data.
- Protein Expression
- Protein Purification
- Binding Affinity
- Aggregation Assessment
- Stability Evaluation
- Developability Assessment
Get told when a licence changes
Licences on these tools do change. One short email when a status flips or a tool is added. No other use of your address.