# What is Open-weight Model?

Also called open-source model, open model.

An open-weight model is one whose trained parameters are published for download, so anyone can run it on their own hardware, inspect it, and adapt it. Open weights are not the same thing as open source, since training data and code are usually withheld and licenses often restrict certain uses. The distinction matters legally and is frequently blurred in coverage.

Publishing weights changes the deployment picture entirely. The model can run inside a private network or an air-gapped environment, it cannot be deprecated or altered by a vendor, its behavior stays fixed until you choose to change it, and it can be fine-tuned or quantized without permission. For regulated data and long-lived systems these properties are often decisive.

Licenses vary widely and have to be read rather than assumed. Some releases use genuine open-source licenses, while others add acceptable-use policies, restrictions on training competing models, or revenue thresholds above which separate commercial terms apply. Calling every downloadable model open source misstates what a good number of these licenses actually permit.

Self-hosting moves cost rather than removing it. Somebody must provision accelerators, size capacity for peak load, apply updates, monitor output quality, and add the safety filtering that a hosted service would have supplied by default. At low or spiky volumes a hosted endpoint is frequently cheaper, while at sustained high volume the economics can reverse.

The capability gap has narrowed but not closed. The strongest open-weight releases now match closed models from roughly the preceding generation on many published evaluations, while the newest closed frontier models generally still lead. Many teams therefore mix both, keeping sensitive or high-volume work in-house and sending the hardest requests to a hosted frontier model.

## Key points

- Weights are downloadable; training data and code usually are not.
- Open weights does not automatically mean an open-source license.
- Enables private, air-gapped, and version-stable deployment.
- Self-hosting shifts cost to infrastructure, operations, and safety tooling.
- The strongest open releases trail the newest closed frontier models.

## In practice

A clinic cannot send patient notes to an external service. It downloads an open-weight model, quantizes it to fit a single on-premises server, and fine-tunes it on de-identified examples of its own documentation style. Notes never leave the building, the model version stays fixed for audit purposes, and the clinic accepts the burden of hosting and monitoring it.

## Related terms

- [Parameters](/en/glossary/parameters)
- [Quantization](/en/glossary/quantization)
- [Small Language Model](/en/glossary/small-language-model)
- [Fine-tuning](/en/glossary/fine-tuning)
- [Benchmark](/en/glossary/benchmark)

[Back to the AI Glossary](/en/glossary)