Nanopath: Fast, Fair, Open Experimentation for Pathology Foundation Models

Abstract

We introduce nanopath, an open-source framework and benchmark for rapidly training and evaluating small-scale pathology foundation models. Using a single H100 GPU, a typical nanopath model trains in a little over an hour on up to one million TCGA pathology tile presentations and evaluates within 25 minutes on a broad suite of tile-level and slide-level tasks. Nanopath enables rapid exploration of research ideas at small scale, so promising training recipes can subsequently be tested with more data and compute. To enable fair comparisons between ideas, we constrain submissions to fixed data and compute budgets while keeping the downstream evaluation protocol fixed. Anyone may clone the codebase, automatically download the data, reproduce or modify existing models, and submit a run to Labless, our public experiment ledger where researchers and AI agents can search through prior attempts. Since its public release in June 2026, nanopath has received 1,269 experiments from 20 contributors. Here, we detail its design and assess how well its rapid evaluation tracks standardized THUNDER, HEST, and PathoBench (CPTAC) benchmarks across six nanopath and 14 reference foundation models. Nanopath's fast evaluation suite strongly tracked these standardized benchmarks, and a few 22M parameter nanopath models even outperformed giant 1.1B parameter models on a subset of downstream tasks. We subsequently discuss how we have since optimized nanopath's evaluation suite to be an even more faithful, quick proxy for external benchmarking. Nanopath provides an accessible testbed for students, researchers, and AI agents to explore pathology foundation modeling through fair, transparent, and directly comparable experiments.

Pathology foundation-model embeddings reveal a morphological landscape of breast lesions beyond diagnostic categories
Y. Muroyama, E. Takaya, T. Suzuki
bioRxiv2 Nov 2026PreprintAll rights reservedFull textGeneral medicinebioRxiv
Generalizing Amyloid β Pathology from Postmortem Human Brains with Vision Foundation Models for Lesion Quantification
Y. Chen, B. D. C. Boon, E. Engelberg-Cook +5
medRxiv5 Oct 2026PreprintAll rights reservedFull textNeurologymedRxiv
Towards reliable morphological subtyping and cancer detection on Barrett's esophagus pathology through ensembling foundation models.
Azar Kazemi, Camillo Saueressig, Jingsong Liu +10
NPJ digital medicine21 Oct 2026Peer-reviewedCC BYFull textGeneral medicineEurope PMCDOAJ
Foundation Model-Enabled Pathology-Informed MRI Model for Preoperative Prediction of Vessels Encapsulating Tumor Clusters in Resectable Hepatocellular Carcinoma: A Multicenter Retrospective Study.
Zhehan Shen, Wanmin Liu, Xiaoming Li +10
Advanced science (Weinheim, Baden-Wurttemberg, Germany)18 Oct 2026Peer-reviewedCC BYFull textMedical imaging & radiologyEurope PMC
Stain Normalisation Does Not Consistently Improve Matched-Location Retrieval with Pathology Foundation Models
Y. Meng, B. Chai, T. Chakraborti
bioRxiv13 Oct 2026PreprintCC BYFull textGeneral medicinebioRxiv

Cite

BibTeX RIS

BibTeX

@misc{scotti2026nanopath,
  title = {{Nanopath: Fast, Fair, Open Experimentation for Pathology Foundation Models}},
  author = {Paul S. Scotti and Hassan Muhammad and Ryan Kim and Nima Ashjaee and Ratna Sagari Grandhi and Saman Rahbar and Raymond Biju and Anish Dulal and Chao Feng and Connor Lane and Benjamin Warner and Tanishq Mathew Abraham},
  howpublished = {arXiv},
  year = {2026},
  month = nov,
  doi = {10.48550/arxiv.2611.10014},
  eprint = {2611.10014},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.10014},
  note = {CC BY 4.0},
}

RIS

TY  - UNPB
TI  - Nanopath: Fast, Fair, Open Experimentation for Pathology Foundation Models
AU  - Paul S. Scotti
AU  - Hassan Muhammad
AU  - Ryan Kim
AU  - Nima Ashjaee
AU  - Ratna Sagari Grandhi
AU  - Saman Rahbar
AU  - Raymond Biju
AU  - Anish Dulal
AU  - Chao Feng
AU  - Connor Lane
AU  - Benjamin Warner
AU  - Tanishq Mathew Abraham
PY  - 2026
DA  - 2026/11/04
PB  - arXiv
DO  - 10.48550/arxiv.2611.10014
UR  - https://arxiv.org/abs/2611.10014
AB  - We introduce nanopath, an open-source framework and benchmark for rapidly training and evaluating small-scale pathology foundation models. Using a single H100 GPU, a typical nanopath model trains in a little over an hour on up to one million TCGA pathology tile presentations and evaluates within 25 minutes on a broad suite of tile-level and slide-level tasks. Nanopath enables rapid exploration of research ideas at small scale, so promising training recipes can subsequently be tested with more data and compute. To enable fair comparisons between ideas, we constrain submissions to fixed data and compute budgets while keeping the downstream evaluation protocol fixed. Anyone may clone the codebase, automatically download the data, reproduce or modify existing models, and submit a run to Labless, our public experiment ledger where researchers and AI agents can search through prior attempts. Since its public release in June 2026, nanopath has received 1,269 experiments from 20 contributors. Here, we detail its design and assess how well its rapid evaluation tracks standardized THUNDER, HEST, and PathoBench (CPTAC) benchmarks across six nanopath and 14 reference foundation models. Nanopath's fast evaluation suite strongly tracked these standardized benchmarks, and a few 22M parameter nanopath models even outperformed giant 1.1B parameter models on a subset of downstream tasks. We subsequently discuss how we have since optimized nanopath's evaluation suite to be an even more faithful, quick proxy for external benchmarking. Nanopath provides an accessible testbed for students, researchers, and AI agents to explore pathology foundation modeling through fair, transparent, and directly comparable experiments.
ER  -

Abstract

Related

Cite