Stanford Benchmarks 32 Foundation Models on Pathology: Vision Models Outperform Vision-Language

iScienceLuvr · x · 2026-08-09

Stanford researchers evaluated 32 foundation models across a diverse suite of 41 pathology tasks. Key findings indicate that pathology-specific vision models (Path-VM) outperform pathology-specific vision-language models (Path-VLM). Additionally, simply scaling model size and training data does not uniformly improve pathology performance, while model ensembling effectively boosts task accuracy.

The top-performing individual models on the benchmark include Virchow2 and Prov-GigaPath (Microsoft), UNI (Harvard), and H-Optimus-0 (Bioptimus).

Original post →

More from Research

Research channel →