Skip to main content
BIO·ATLASby Bayes Pharma.ai
Search
AccountStart evaluation
AI drug discovery models/Tissue, Pathology & Imaging/GenBio-PathFM
model-family passport · Review date not recorded

GenBio-PathFM

A 1.1B-parameter histopathology foundation model trained with morphology-aware curation and dual-stage learning.

4/7Evidence fields documented
60-SECOND EVALUATION VIEW

What should a scientist know before using GenBio-PathFM?

SupportedEvidence supports the stated context with explicit boundaries
Best suited forRepresentation · Prediction
Evidence supportsTHUNDER / HEST / PathoROB: Technical report / open evaluation
Evidence does not establishUniversal superiority, therapeutic success, clinical utility or regulatory acceptance.
Major limitationPerformance depends on the evaluation dataset and operating conditions.
CompareAdd to evaluationView evidence
Current registry recordVersion history not yet curated1 recorded release · Review date not recorded. A newer version is not assumed to be universally better.

What it is

GenBio-PathFM is a histopathology foundation model built from public data with morphology-aware curation and a dual-stage JEPA-plus-DINO learning strategy. BioAtlas records benchmark claims as publication/developer evidence rather than a universal pathology ranking.

Evidence trail

BioAtlas keeps the path from source to decision visible. A connection records provenance; it does not imply that evidence is sufficient for every context.

Sources4 connectedPrimary resources and normalized claims
Claims1 normalizedIntegrated discovery platform
EntityGenBio-PathFMmodel-family · Version history not yet curated
ReviewReview date not recordedReview date not claimed
ConclusionContext requiredAdd to an evaluation before operational use

Model passport

Entity typemodel-family
OrganizationGenBio AI
Model family introduced2026
AccessOpen source
Commercial useAllowed / verify checkpoint terms
DeploymentSelf-hosted
ComputeGPU recommended
Domainspathology
Biology → representation → computation → evidence

How GenBio-PathFM represents biology

model-familypathology

Category is navigation. These fields describe the model-specific computational transformation and deliberately override broad category defaults.

1 · Biological inputs
Whole-slide images or pathology tiles
2 · Input representation
Image patches / tiles
3 · Internal representation
Morphology-aware tissue embeddings
4 · Architecture
Vision foundation modelJEPA + DINO dual-stage learning
5 · Learning objective
Self-supervised pathology representation learning
6 · Output representation
Dense vectorsLabels / scores

Biological scale

tissuewhole-slide

Modalities & tasks

Histopathology imageRepresentationPredictionRetrieval

Registry, claims and frontier intelligence

Versioned registry

Version history not yet curated

1 version record · release year not yet normalized. Model-family identity remains separate from capability and access changes.

Explore version lineage →

Inputs and outputs

Inputs

Whole-slide images or pathology tiles

Outputs

Pathology embeddingsTask predictions

Scientific and technical profile

Scientific principles

Morphological diversity curationSelf-supervised pathology learning

Technology

1.1B parametersJEDI (JEPA + DINO)

Evaluation evidence

Dataset or evaluationTHUNDER / HEST / PathoROB
Task or metricHistopathology representation and downstream performance
Evidence statusTechnical report / open evaluation
Open source ↗

Task-specific evidence only; not comparable as a universal leaderboard score.

Integrated discovery platform

THUNDER / HEST / PathoROB

Version history not yet curated · Split details not yet normalized
developer-reported

A structured benchmark claim is recorded; consult the linked source for numeric values and protocol details.

Claim caveats
  • Protocol, split and implementation details must match before comparing this claim with another result.
Open benchmark claim →

Known limitations

  • Performance depends on the evaluation dataset and operating conditions.
  • Task-specific benchmark results should not be compared across unlike domains.
  • Outputs require task-specific scientific and experimental validation.

Milestones

2026

Reported evaluations include THUNDER, HEST and PathoROB; cross-paper comparability remains task-specific.