Model Serving & Economics
Closed model
A model whose weights and core internals are not publicly released by its provider.
Example
Customers use a model through an API but cannot inspect or download its weights.
Why it matters
It lets providers offer a managed AI service while retaining control over the underlying model.
You'll hear it
“We can test its answers, but we can't inspect everything inside.”
Can you control it?
Sometimes
Sometimes. Some products or workflows let you influence a closed model, while others keep it fixed or manage it automatically. Check the settings and documentation for the tool you are using.
Common questions
- Does closed mean unsafe?
- No. Safety depends on behavior, safeguards and use; public access to the model is a separate question.
- Can customers adapt its behavior?
- Sometimes. Providers may offer custom instructions or training options without releasing the model itself.
- What happens if the provider withdraws it?
- You may need to switch models. Keep copies of your own instructions, tests and application data.