A base model is a trained model used as the starting point for later specialization or deployment. Developers may fine-tune it, add retrieval, give it tools or wrap it in a larger system designed for a particular task.
Separating the base model from the full system matters when evaluating results. Performance may come from prompts, test-time methods, data selection, tools and human decisions rather than from the unchanged model alone.
ELI5
A base model is the starting trained AI model before developers adapt it for a particular product or task. They can add special training, tools, retrieval, prompts or other system parts around it.
For example, the same base language model could become a support assistant after being connected to approved documents and ticket tools. When judging the result, it matters whether the improvement came from the base model itself or from the added data, tools and workflow.
