OpenFold3 production run cuts antibody-antigen inference ~5x

OpenFold3 is an open-source structure prediction model, and MindWalk says it chose that route deliberately — the ability to predict a protein's shape is increasingly something any team can buy, while the value of knowing what that shape means for a disease and for programs already run is where its own layer, ReefIQ™, sits. The deployment matters to that layer because it expands the compute available beneath it, coming after ReefIQ's launch and as MindWalk says it is scaling enterprise adoption.

On the engineering side, the claim is about turning the computing into a repeatable path rather than a one-off build: the company frames the result not as a benchmark win, but as a faster, cheaper way to onboard a partner, so that a partner's programs can start running in minutes instead of waiting on a custom setup. MindWalk also describes the exercise as an independent validation of AMD Inference Microservices for OpenFold3 in a production life-sciences setting.

The test design is worth noting. MindWalk says it evaluated the deployment around its own production workloads and ran them under its own governance and human scientific review, rather than on a set chosen to look good — which is the standard its partners apply, per the company. Whether the reported gain translates across the full range of prediction types the company describes, beyond antibody-antigen work, is likely to be the question that determines how broadly the result is read.