Model reference · open weights

CONCH

CONCH is an open-weight embedding model from MahmoodLab. CONCH (BF16) weighs 802 MB; the smallest configuration that runs it is RTX 3060 12 GB.

• CONCH is a vision-language foundation model for histopathology developed by MahmoodLab.
• It is designed for image feature extraction and supports tasks such as classification, retrieval, and segmentation.
• The model consists of a ViT-B/16 vision encoder with 90 million parameters and a text encoder with 110 million parameters.
• It was pretrained on 1.17 million image-caption pairs and is released under the CC-BY-NC-ND-4.0 license for non-commercial academic research.

Summary of the MahmoodLab/CONCH model card, 2026-10-01

Hugging Face ↗