> ## Documentation Index
> Fetch the complete documentation index at: https://docs.biohub.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# fold_max_accuracy

> Run maximum-accuracy structure prediction.

Use [Batch Submit](/api/protein/models/batch_submit) with `endpoint: "fold_max_accuracy"` and one or more inputs in `payload`.
Each item must name `esmfold2-2026-05-cutoff-2025` as its model.
This operation is available on Biohub's Batch API.
The schema below describes one payload entry. See the [Batch API overview](/api/protein/models/batch_overview) for submission, polling, and downloads.

## Inputs and limits

Supply a molecular complex through `all_atom_input` with protein, RNA, DNA, or ligand entries.
The total input length is limited to 4096 tokens.
Protein residues and RNA/DNA nucleotides count as one token each; ligands and modified residues count by heavy atom.

The service generates its own MSA by searching UniRef30 and ColabFold with the following settings:

* `max_prefilter` = 100,000
* `max_accept` = 10,000
* `diff` = 3,000
* `expand_diff` = 3,000

It then fans out to 5 folds using 5 different seeds and selects the best predicted structure.
The supported output options are `include_pae`, `include_pair_chains_iptm`, and `include_embeddings`.