AI for Bio · Model

IntelliFold-2

Hong Kong co-folding model billed as an AF3 improvement; Pro tier is server-only

Structure & co-foldingAI-native biotechHong KongActive中文

Overview

NameIntelliFold-2 · aliases: IntelliFold-2, IntelliFold-v2
Full nameOfficial materials use IntelliFold-2 only; no expanded name
Ambiguous names (not auto-linked)IntelliFold, IntFold, IntFOLD
FamilyNot a family entry. Covers IntelliFold-2 (Flash/standard/Pro) only; IntFold counted as predecessor
OrganizationIntelliGen AI
Organization typeAI-native biotech
CountryHong Kong
Model classCo-folding (protein/nucleic acid/small molecule)
DomainStructure & co-folding (also: Antibodies)
First release† partly verified · code release
Basis: GitHub release v2.0.0 (the same month's bioRxiv report was not date-checked)
Note: The predecessor IntFold release of 2025-07 is not this record's first release source
Other milestones
  • First public evidence: † · GitHub release v2.0.0 source
  • Preprint v1: † · bioRxiv technical report (release notes) source
  • Code release: † · GitHub release v2.0.0 source
  • Weights release: † · Hugging Face card published the Flash and standard weights source
Latest update · code release
Basis: GitHub release v2.0.4
v2.0.3 moved inference onto the AlphaFold 3 JAX engine (HF card updated 06-15) and v2.0.4 followed on 06-23; later commits are steering fixes and maintenance source
The most recent sourced event, not a claim that nothing changed since.
StatusActive · Tagged releases, with commits to 2026-10 as maintenance
OpennessCode: Open · Apache-2.0
Weights: Open · Apache-2.0
Flash and standard code and weights are Apache-2.0 and commercial-use; the Pro tier is server-only and not reproducible
Benchmarks reported2 counted (self 1, modified protocol 1)
Wet-lab validationNone recorded
Unverified fields† release_milestones.paper_v1.date, background.funding
As of2026-11-16

What it does

Predicts co-folded complexes of proteins, nucleic acids and small molecules, positioned as an improvement on the AlphaFold3 architecture. Flash and standard builds are public while the Pro tier ships only with the hosted service, combining open weights with a commercial offering. It sits in the same open AF3-reproduction line as Protenix and SeedFold.

Benchmarks reported

Every row is one pairing from the survey's relation file; a score belongs to the benchmark page's ranking, where it is shown with metric, eval set, source type and date. 2 of 2 pairings are counted.

BenchmarkRelationEval set / protocolSource typeNote and evidence
31 FoldBenchDeveloper report, official protocolFoldBench antibody-antigen / protein-ligandSelf-reportedPublished as a figure only; the predecessor reports numbers · evidence
05 PoseBustersDeveloper report, modified protocolPoseBusters V2 (README, steering option)Self-reportedApproximate values in the repository README with a steering option; no table †partly verified · evidence

Other evaluations (outside the 44)

  • Predecessor IntFold self-report (self report) Self-reported: The 2025-07 report self-reported protein-protein and antibody-antigen results on FoldBench; that is the predecessor's protocol and the two generations must not be mixed
  • Controllable adapters and ranking (capability claim) Self-reported: The predecessor paper also claims allosteric-state, constraint and affinity adapters plus a training-free similarity ranking (self-reported)

Developers report: Developers report gains over AlphaFold3 on FoldBench antibody-antigen and protein-ligand tasks; the release notes give no citable table (self-reported).

Independent evidence: No independent evidence found. No third-party rerun or board entry found

Training data

Group in Figure 7: Public data only.

ClassValueDetail
a. Public databasesYesReprocessed PDB plus self-distillation data; details not reported source
b. Private or licensed dataNot disclosed
c. Own wet-lab data used for trainingNo
d. Wet lab used only for validationNo

Wet-lab validation

None. The developers report no wet-lab validation.

Background

  • Organization: IntelliGen AI is headquartered in Hong Kong and, per media reports, was co-founded in 2024-10 to compete with big-lab drug discovery; headcount is undisclosed (second-hand).
  • Funding: Media report an angel round of tens of millions of dollars; neither amount nor investors are official (second-hand, unverified).
  • Products and partnerships: Flash and standard weights are public and the Pro tier is offered through the server; no pharma partnership appears in public material.
  • Related models: AlphaFold 3, Protenix, SeedFold

Reading notes

  1. The headline claim sits on FoldBench but has no citable table, so it stays a vendor claim.
  2. The PoseBusters line exists only as README approximations under a different protocol.
  3. IntFold and this generation use different protocols; do not read their results together.

Caveats

  1. The Pro tier is closed, so its best server-side results cannot be reproduced.
  2. No official Runs N' Poses report was found, so that pairing is treated as absent.
  3. Company and funding details come from media reports, not official confirmation.

Sources

  1. IntelliFold-2 bioRxiv technical report (preprint) · †
  2. IntelliFold GitHub releases (repository)
  3. Hugging Face model card (repository)
  4. Predecessor IntFold technical report (preprint)
  5. Press report on the company (press)

Corrections

1 correction
FieldWasNowBasisFound
latest_update2026-06-16 (v2.0.3)2026-06-23 (v2.0.4 is the later tagged release)source2026-11-16

Part of the AI for Bio benchmarks survey · all 112 models · open in the explorer. Record as of 2026-11-16.