Bioptimus

Bioptimus is an AI-first biotech company building foundation models intended to represent human biology across modalities and scales, from cells and tissues to organs and patients. Its first released model, H-Optimus, is a foundation model for histology (H-Optimus-0, described as a 1.1 billion parameter model for histology images, followed by H-Optimus-1). In December 2025 the company unveiled M-Optimus, which integrates pathology, spatial transcriptomics and genomics into a single architecture and is positioned as a "world model of biology" producing a unified embedding space across tissue, molecular and clinical levels.

Applications marketed by the company include drug target and biomarker discovery from routine H&E histology slides (for example predicting the spatial distribution of ADC targets, immune microenvironment and stromal architecture), indication expansion by scoring archival trial cohorts, diagnostics and spatial biology pipelines combining histology, spatial transcriptomics and genomics, and treatment-response prediction and clinical trial design, including in-silico readouts such as objective response rate and progression-free survival and synthetic trial arms. The company states its models can be fine-tuned to predict gene expression from multimodal embeddings of slides, routine sequencing and spatial transcriptomics.

Alongside models, Bioptimus runs STELA (Spatial Tissue Embedding Learning Atlas), a multi-institutional data-generation initiative assembling deeply profiled, clinically linked multimodal patient data. STELA aims to profile up to 100,000 patient specimens across a global research network, described by the company as roughly a 20x increase over existing available data, and is built through partnerships with data holders and technology providers.

Founding story

Bioptimus was founded in 2024 by six co-founders, several of whom came from Owkin and from large-scale machine learning research groups. CEO Jean-Philippe Vert was previously Chief R&D Officer at Owkin and a research lead at Google Brain; CTO Rodolphe Jenatton was a senior research scientist at Google Brain after five years at Amazon; VP of Research Zelda Mariet worked at Google DeepMind after a PhD at MIT; VP of AI Felipe Llinares-López worked on protein and DNA language models at Google DeepMind. According to Vert, building biology foundation models was not part of Owkin's roadmap, and a separate entity was better placed to assemble the data volume, compute and modality breadth required, with Owkin supporting and partnering with the new company and its multimodal patient data used in training.

Business model

Bioptimus develops foundation models and licenses or provides access to them for biomedical research and pharmaceutical development. Models are made available directly (early/pioneer access programmes for M-Optimus, access requests, open release of H-Optimus-0) and via cloud distribution, including a partnership with Amazon SageMaker AI for H-Optimus-1. In parallel it operates STELA, a collaborative data engine in which partner institutions contribute clinically linked multimodal patient data and in return benefit from the resulting atlas.

The company offers commercial access to its foundation models for pharmaceutical and biotech development pipelines, with tiered availability including open/benchmark pathology models and limited early access to M-Optimus; specific pricing terms are not disclosed.

Traction

As of the company website: over 1,184,533 total model downloads, more than 100 papers published using its models, use by 16 of the top 20 pharma companies and more than 1,000 clinical practices/institutions. Published case studies cover breast cancer recurrence risk prediction with H-Optimus-1 and STAMP, pathology report generation by ICGI researchers, spatial transcriptomics workflows in IBD research, and a clinical cancer prediction study with MIT. STELA data partnerships include 10x Genomics and Broad Clinical Labs, FFCD (clinical trial data from over 3,000 gastrointestinal cancer patients) and Cancer Research Horizons' S:CORT colorectal cancer dataset (over 1,750 patient profiles, the first UK cohort in STELA).

Latest developments

In August 2026 Bioptimus announced a partnership with Cancer Research Horizons for access to the S:CORT colorectal cancer dataset, adding over 1,750 patient profiles and the first UK cohort to STELA. Earlier in 2026 it signed a partnership with FFCD covering data from more than 3,000 gastrointestinal cancer patients (June 2026) and launched STELA with 10x Genomics and Broad Clinical Labs (March 2026), targeting up to 100,000 profiled specimens. In December 2025 it unveiled M-Optimus and expanded access to H-Optimus-1 through Amazon SageMaker AI.