IT lexicon AI & ML Model distillation

Model distillation

AI & ML På svenska → Updated: 2026-05-23

Train a smaller, faster "student" model on the output of a large "teacher" model.

Big models (GPT-4, Claude Opus) are expensive to run. Distillation squeezes most of the knowledge into a 10x smaller model. The backstory behind Haiku, Mini, Flash variants.

← Back to the lexicon