Biohub
Publisher of the ESMFold-2, ESMC, and ESM3 open protein model lines.
Models on Rafflesia
ESMFold-2 Fast
Predict one small protein structure directly from sequence with the frozen ESMC 6B backbone and the inference-optimized, non-MSA ESMFold-2 Fast checkpoint.
ESMFold-2
Predict protein monomer and complex structures directly from sequence or an optional MSA with ESMFold-2.
ESMC 300M
Create pooled or residue-level protein sequence representations with the compact ESM Cambrian model.
ESMC 300M
Score amino acid substitutions as ESM Cambrian masked marginals with the compact 300M checkpoint.
ESMC 600M
Create higher-capacity pooled or residue-level protein sequence representations with ESM Cambrian.
ESMC 600M
Score amino acid substitutions as ESM Cambrian masked marginals with the 600M checkpoint.
ESMC 6B
Create pooled or residue-level protein representations with the 6B-parameter ESM Cambrian checkpoint used as the frozen language-model backbone of ESMFold-2.
ESMC 6B
Score amino acid substitutions as ESM Cambrian masked marginals with the 6B-parameter checkpoint.
ESM3 1.4B Open
Create pooled or residue-level protein representations with the open 1.4B-parameter ESM3 checkpoint.
ESM3 1.4B Open
Predict a protein structure by decoding the ESM3 structure track from sequence with the open 1.4B-parameter checkpoint.
ESM3 1.4B Open
Sample protein sequences conditioned on a fixed backbone by decoding the ESM3 sequence track with the open 1.4B-parameter checkpoint.
ESM3 1.4B Open
Complete a partially specified protein across the sequence, structure, secondary structure, solvent accessibility, and function tracks with the open 1.4B-parameter ESM3 checkpoint.
ESM3 1.4B Open
Score amino acid substitutions with ESM3 masked marginals from the open 1.4B-parameter checkpoint.
About
Biohub publishes open-weight protein models in the ESM lineage. Rafflesia identifies callable variants by their upstream names: ESMFold-2 and ESMFold-2 Fast, ESMC at 300M, 600M, and 6B parameters, and the open 1.4B ESM3 checkpoint.
Each model is called through a stable provider/family/variant endpoint id under the runs API.
Use these models
Every endpoint speaks the same run lifecycle: POST the endpoint id and a typed input, get a run back with outputs, provenance, and measured billing.
# one run lifecycle, every endpoint
curl https://api.rafflesia.ai/v1/runs \
-H "Authorization: Bearer $RAFFLESIA_API_KEY" \
-H "Content-Type: application/json" \
-d '{
"endpoint": "biohub/esmfold/2-fast/fold",
"input": { "sequence": "MKTAYIAK..." }
}'