"World models" in biopharma sound like a frontier that's years away. We mapped 69 of them. Two-thirds are already deployed or partnered.
‍
We defined world models strictly: systems that simulate biological processes, disease trajectories, intervention responses, or manufacturing operations across interacting variables. No single-task predictors. No biomarker classifiers. Even with that bar, the landscape is far bigger than the conversation suggests.

The maturity pattern is what matters. Manufacturing digital twins lead, with 10 of 16 efforts already in production. Process physics and sensors are observable in ways biology is not. This is where world models proved the concept works.

Biology simulation is the largest domain at 21 efforts, but mostly partnered rather than deployed. The window to own a defensible platform here is still open, and the companies racing to close it are AI-native biotechs like Recursion, Bioptimus, and CytoReason. Not large pharma.

Large pharma licenses foundation models from external partners but builds world models internally, clustering in QSP and manufacturing where regulatory returns are already clear. No large pharma company in our dataset is attempting a broad biology world model. Those capabilities will enter pharma through acquisition.

The category is still being defined. The competitive advantage right now is not just building a world model but shaping what the term means.