Building & Running AI intermediate

Batch Inference

Also called: offline inference

Running many predictions together as a job instead of one at a time in real time.

Batch inference processes a queue of inputs where latency does not matter, using hardware far more efficiently. Providers often discount it substantially. If a human is not waiting for the answer, it is the cheaper path.

In practice: Classifying a year of support tickets overnight at half price.