jeff vs Jev vs Jevlike vs Von

Pick jeff if you are already using the Jev API and need to move inference on-premises, whether for data-residency, cost, or offline requirements, and you are comfortable running it from source.

Pick Jev if you want a production-ready, managed decision API with a free tier and can accept a 70–500ms round-trip latency, and you do not want to operate any infrastructure yourself.

Pick Jevlike if your use case involves a list of options that changes frequently and you want to train a small, task-specific model that handles selection in a single forward pass rather than calling any external service.

Pick Von if you need discrete, probabilistic, or ordinal inference to run locally under 25ms and you want a model that is not tied to the Jev API surface or its encoder architecture.