Glossary
Open-weight models
Models whose weights are freely available and which can therefore be run yourself, in your data centre or your own cloud.
The appeal is control and data protection: requests never leave your environment, and nobody changes the model underneath a running application.
The price is operations: GPUs, scaling, updates, and the realization that idle time is paid for here while it is free with a provider.
How you notice it
- Data must not leave your own network.
- Vendor lock in is a stated risk.
- There are uniform tasks in high volume.
Frequently asked
Does self hosting pay off?
Only above a certain volume. GPU hours run around the clock, API calls only on use. Below a few million requests a month the provider is usually cheaper. Data protection and independence can still tip the decision.
