French AI startup Bioptimus said it is building a foundation-model approach to predict spatial gene expression and downstream tumor and patient behaviors from H&E tissue slides and transcriptomic data. The company, emerging from stealth in 2024, received $35 million in seed funding led by Sofinnova Partners and later raised $41 million in a round led by Cathay Innovation. Bioptimus’ platform—marketed as a “world model” for biology—starts with H-Optimus models that take routine pathology images as input and output predictions spanning gene mutations, protein expression, and recurrence risk. A subsequent model, M-Optimus, combines slide images with bulk RNA sequencing to improve spatial gene-expression inference. The report noted that H-Optimus-0 is open-source, while H-Optimus-1 is free for academics but requires commercial licensing, and that large pharmaceutical companies use the approach. The company said it now has about 35 employees and is expanding model capabilities toward multimodal biology tasks.