Models

Phonos runs published models built by research groups and companies. Each entry says what the model takes, what it returns, how to read its confidence, and who built and serves it.

Provenance reads built by · served by. Models served through NVIDIA BioNeMo NIM process your input on NVIDIA infrastructure, which may be in the United States — see the privacy policy. NVIDIA’s hosted models are used only in the Research Preview, for testing; paid plans run on machines operated by FyreFly Systems.

Open Phonos ↗

Boltz-2

Biomolecular foundation model for structure and affinity.

Predicts 3D structures of complexes — proteins, nucleic acids, small molecules and their combinations — and estimates binding affinity for a ligand.

MIT Jameel Clinic · NVIDIA BioNeMo NIM

Inputs, outputs and limits

RFdiffusion

Diffusion model for protein backbone design.

Generates new protein backbones step by step from noise: binders against a target, scaffolds around a motif, or proteins from scratch. ProteinMPNN then designs sequences for them.

Institute for Protein Design, University of Washington · NVIDIA BioNeMo NIM

Inputs, outputs and limits

MolMIM

Generative model for controlled small-molecule design.

A probabilistic auto-encoder that learns a clustered latent space over SMILES strings. MolMIM generates new small molecules with desired properties under constraints, using CMA-ES-guided sampling to optimise QED or logP.

NVIDIA · NVIDIA BioNeMo NIM

Inputs, outputs and limits

Evo 2

DNA foundation model across all domains of life.

A genomic language model with 40 billion parameters, trained on nearly 9 trillion nucleotides from bacteria, archaea, plants and humans, with up to 1 million base pairs of context. On Phonos it generates DNA that continues your sequence.

Arc Institute · NVIDIA BioNeMo NIM

Inputs, outputs and limits

At a glance

Eight models available, three retired
Model	Family	Takes	Returns	Status
Boltz-2	Structure and complexes	Sequence; ligand, optional	Structure (mmCIF), affinity	Available
OpenFold3	Structure and complexes	Sequence; MSA and ligand, optional	Structure (PDB, mmCIF)	Available
DiffDock	Docking	Protein structure; ligand	Poses (SDF)	Available
RFdiffusion	Protein design	Target structure; contigs	Backbones (PDB)	Available
ProteinMPNN	Protein design	Backbone (PDB)	Sequences (FASTA)	Available
MolMIM	Small molecules	Seed molecule (SMILES)	Molecules (SMILES)	Available
GenMol	Small molecules	SMILES or masked SAFE	Ranked molecules (JSON)	Available
Evo 2	DNA and RNA	DNA sequence	Nucleotides (JSON)	Available
ESMFold	Structure	Protein sequence	Structure (PDB)	Retired
AlphaFold2	Structure	Protein sequence	Structure (PDB)	Retired
AlphaFold2-Multimer	Complexes	Protein sequences, one per chain	Complex structure (PDB)	Retired

By task

One page for each kind of work, with inputs, outputs, limits and provenance for every model.

Structure and complexes

Predict how a chain folds, alone or together with a ligand.

Boltz-2, OpenFold3 See the models

Docking

Place a small molecule into a protein.

DiffDock See the model

Protein design

Generate backbones, then design sequences for them.

RFdiffusion, ProteinMPNN See the models

Small molecules

Generate and optimise drug-like molecules from a seed.

MolMIM, GenMol See the models

DNA and RNA

Generate DNA. Boltz-2 and OpenFold3 also predict DNA and RNA structures.

Evo 2 See the model

Retired

NVIDIA retired the hosted endpoints of these models in 2026. They no longer run on Phonos; for structures and complexes use Boltz-2 or OpenFold3.

ESMFold — Retired · endpoint closed on 25 August 2026. Predicted a protein structure from a single sequence, without an alignment. Meta AI (FAIR).
AlphaFold2 — Retired · endpoint closed on 24 September 2026. Predicted a single-chain protein structure from its sequence and an alignment. Google DeepMind.
AlphaFold2-Multimer — Retired · endpoint closed on 24 September 2026. Predicted the structure of protein complexes from several chains. Google DeepMind.

Models

Boltz-2

At a glance

By task

Structure and complexes

Docking

Protein design

Small molecules

DNA and RNA

Retired

Cookie settings