# What is model distillation?

Distillation trains a smaller, cheaper model to imitate a larger one by learning from its outputs. Teams use a frontier model to handle a task, collect its best responses, then distill them into a small open model that serves the same task at a tenth of the cost and latency.

Distillation is the standard graduation path: prove the workflow on a frontier API, then distill onto owned open weights once volume justifies it.

Combined with your own outcome data, distillation plus fine-tuning often beats the original teacher on the specific task, because the student specializes.

## Related terms

- [Fine-tuning](https://allocate.network/glossary/fine-tuning.md)
- [Open-weight models](https://allocate.network/glossary/open-weights.md)
- [Training signal](https://allocate.network/glossary/training-signal.md)

---

[HTML page](https://allocate.network/glossary/model-distillation) · [Machine-readable catalog](https://allocate.network/catalog.json)