Benchling as AI Lab Control Plane

Model Hub turns Benchling from a system of record into a system of execution for scientific AI. Instead of exporting sequences into separate modeling tools, waiting for an ML team to wire up GPUs, then pasting results back into a notebook, scientists can pick a registered protein or molecule inside Benchling, run a model from a shared catalog, and have the prediction land back on the same experimental record with permissions, timestamps, and lineage intact.1 2 3 4 5 6

• In biologics this is especially valuable because the input is already structured. Benchling Biologics stores antibody sequences, formats, lots, assay results, and lineage in one antibody aware data model, so a scientist can run developability or structure models on registered candidates instead of rebuilding inputs by hand in another system.4 6 7
• The catalog is not generic AI chat. Model Hub already includes structure models such as AlphaFold 2, Chai-1, Boltz-2, OpenFold 2, OpenFold 3, and Protenix, plus property prediction models like Lilly TuneLab and Inductive Bio. That makes Benchling a distribution layer for specialized scientific models, not just a lab notebook with an AI assistant attached.2 5 8
• The commercial logic is sticky. Benchling already sits on top of experimental records for more than 1,200 customers and roughly $210M ARR as of mid 2024. Once model runs, outputs, and audit trails also live there, switching away means replacing both the underlying R&D system and the attached AI workflow layer.9 10 11

This is heading toward a world where Benchling is the control plane for AI native lab work. As more partner models and bring your own model options arrive, the winning product will be the one that keeps proprietary lab data, model execution, and experimental follow through in one place, and Benchling is building directly toward that position.2 3 5 8