MindWalk Cuts Antibody-Antigen Inference Time About 5x in AMD-Vultr OpenFold3 Deployment

OpenFold3 moves from model availability into MindWalk’s production workflow

MindWalk Holdings says it has put the open-source OpenFold3 structure-prediction model into a production environment using AMD Inference Microservices on Vultr Kubernetes Engine, with compute provided by AMD Instinct MI325X GPUs.

The September 21 update is notable because the test covered the full production path rather than a narrow benchmark. MindWalk ran preprocessing through inference inside its discovery workflows across protein, RNA, DNA, ligand and antibody-antigen prediction under the company’s governance and human scientific review.

That gives MindWalk a standardized route for bringing a sophisticated biological model into day-to-day discovery work, rather than treating model deployment as a separate engineering project each time a new enterprise program begins.

Source: [1]

The headline result is a roughly fivefold inference improvement

In MindWalk’s evaluated antibody-antigen workloads, inference time fell by approximately five times compared with the company’s prior environment. MindWalk also said a production-grade prediction environment could be deployed in minutes using the AMD and Vultr stack.

For a life-sciences platform, speed at this layer matters because structure prediction can sit inside a much larger chain of target analysis, candidate evaluation and experimental work. Reducing infrastructure setup and inference time can let scientists cycle through more questions without rebuilding the computing environment around each program.

The company deliberately evaluated the system using its own production workloads, making the result directly relevant to the way ReefIQ is being used as MindWalk expands its enterprise-facing platform.

AMD and Vultr give ReefIQ a more elastic compute foundation

MindWalk describes ReefIQ as a biological context layer that reconnects discovery data into governed, queryable information for AI models and agents. The new deployment expands the compute available beneath that platform using Kubernetes-native infrastructure and cloud GPUs that can be provisioned as workload needs change.

That architecture is a practical fit for enterprise onboarding. A new partner can bring a different mix of proteins, antibodies or other biological workloads, while MindWalk can scale the underlying compute without creating a bespoke stack from scratch.

The company has already been building around AMD Instinct technology and publicly demonstrated ReefIQ on AMD hardware earlier in 2026. The OpenFold3 deployment takes that relationship a step further by showing a complete production workflow running on the infrastructure.

Source: [1]

The infrastructure work supports MindWalk’s commercial platform push

MindWalk commercially launched ReefIQ in June 2026 and said in its latest quarterly update that it was in active discussions with multiple large pharmaceutical companies on enterprise deployments. The company has been positioning its biological data layer and LensAI reasoning tools as the context around increasingly accessible foundation models.

The September deployment supports that strategy by making the model-and-compute layer faster to stand up. MindWalk’s differentiating argument is that the value sits in the biological representation and context accumulated around the models, while open models and standardized infrastructure can be swapped or scaled as technology evolves.

Putting OpenFold3 into production on an AMD-Vultr stack gives that model-agnostic architecture a concrete operating example and gives the company more capacity for additional enterprise programs.