Batch folding

Use Batch Submit with endpoint: "fold" for protein sequences or endpoint: "fold_all_atom" for complexes containing proteins, RNA, DNA, and ligands. Supported models: esmfold2-2026-05 and esmfold2-fast-2026-05. Each task predicts one structure with your folding configuration. Unlike fold_max_accuracy, these endpoints do not generate an MSA automatically or select the best of multiple predictions. You can supply an MSA to esmfold2-2026-05; the fast model ignores MSAs.

Fold protein sequences

import os

from esm.sdk.api import FoldRequest, FoldingConfig
from esm.sdk.forge import ForgeBatchClient

with ForgeBatchClient(
    token=os.environ["ESM_API_KEY"], model="esmfold2-fast-2026-05"
) as client:
    submission = client.fold.submit(
        requests=[
            FoldRequest(sequence="MKTAYIAKQRQISFVKSHFSRQ"),
            FoldRequest(sequence="GSHMASMTGGQQMGRGS"),
        ],
        config=FoldingConfig(include_pae=True),
    )
    print("Batch ID:", submission.batch_id)
    client.fold.wait(submission)
    proteins = client.fold.results(submission)

Each successful result is an ESMProtein with predicted structure and confidence metrics. Use | to separate protein chains within a sequence. For molecular complexes, use client.fold_all_atom with FoldAllAtomRequest; successful results are MolecularComplexResult objects. See Fold all atom for input construction. Both handlers also provide async_submit, async_wait, async_results, and async_run. Save the returned batch ID before waiting so you can recover results later. See the Batch API overview for failures, cancellation, and result retention.

Protein Models

Batch API

ESM Atlas

Fold protein sequences

Protein Models

Batch API

ESM Atlas

​Fold protein sequences

Fold protein sequences