Checkpoint

A checkpoint is a saved snapshot of an AI model at a particular point in training.

Think of training an AI model like writing a long document with autosave. A checkpoint is a saved copy of the model's learned settings at one stage. Developers can keep several checkpoints to compare results, recover from a bad training run, or choose a version that works best for a task. For people using AI tools, checkpoints matter because the version behind a tool can affect its answers, speed, and behavior. For example, a team might train a model for days and save a checkpoint each day, then test which saved version gives the most useful responses.