Benchling moves scientific AI models into the R&D record: structure predictions stop landing on a shared drive
Synopsis
Benchling's Model Hub lets scientists discover, run and track scientific AI models inside Benchling, with each prediction's inputs, model version and timestamp wired back into the R&D record. Previously, running a model meant waiting for DevOps to stand up infrastructure, and results landed on a shared drive disconnected from the sequence that produced them. The hub currently carries structure prediction models including AlphaFold 2, Chai-1, Boltz-2, OpenFold 2, OpenFold 3 and Protenix, plus property prediction models such as Lilly TuneLab, Inductive Bio and open-source DeepSP, with batch prediction and a v3 alpha API.
Interpretation
Running scientific AI models and recording their results now happens on the same platform where the R&D data lives, so a prediction stays tied to the experiment that produced it instead of scattering across tools. Previously a scientist who found a promising model waited weeks for DevOps to set up supporting infrastructure, and the result landed on a shared drive disconnected from the sequence that produced it, forcing the next scientist to start from scratch. Benchling says every Model Hub prediction returns to the R&D record with its inputs, model version and timestamp, and cites its 2026 Biotech AI Report finding that protein structure prediction was the second most widely adopted AI use case in R&D at 71%.
A user opens the hub from a navigation-bar icon, picks inputs and a model, and gets structured results with a full audit trail, with no code and no GPU setup. Model Hub puts the model catalog, running and tracking in one place, replacing the hunt across point solutions. The hub currently includes structure prediction models AlphaFold 2, Chai-1, Boltz-2, OpenFold 2, OpenFold 3 and Protenix, along with property prediction models such as Lilly TuneLab's proprietary developability models, Inductive Bio for small molecules and open-source DeepSP.
Perspective
The report covers the product capabilities Benchling describes: the model catalog, no-code runs, audit trails, batch prediction, GPU-accelerated alignments, faster structure runs and the v3 alpha API, plus which structure and property prediction models are included. It is aimed at scientists and computational teams running scientific AI models in R&D. Generative protein design models and other model types are described as on the way and are not yet listed.
The adoption figure comes from Benchling's own 2026 Biotech AI Report, and the material does not state that survey's sample or methodology. The material says running enough structure predictions against a team's own proteins builds a dataset that can calibrate and improve later models, but gives no measurement of that effect. Generative protein design models are listed as on the way with no timing.