What is a closed-source model?
A closed-source model is an AI model that its developer keeps private. The weights are never released, the training data and methods are described only in general terms, and the model is available solely through the developer’s own service, usually an API billed per token or a subscription product. The models behind ChatGPT and Claude are examples. The customer sends a prompt over the network, the provider’s servers run the model, and the answer comes back.
Why closed-source models matter
Closed models are often the most capable available at any given time, and they need no hardware or operations from the customer. The trade-offs are control and data handling. Every prompt, including any customer record or contract pasted into it, leaves the company’s environment and is processed under the provider’s data terms. The provider can change the model, retire a version, or alter pricing on its own schedule, which can shift the behavior of an application without the company changing anything. A professional services firm that built a review workflow on one version may find its outputs formatted differently after an update.
What to watch for
- Get the data retention and training-use terms in writing, and confirm whether an enterprise agreement differs from the public terms.
- Pin to a dated model version where the API allows it, and re-test before moving to a new one.
- Use the OpenAI-compatible API format so a switch to another model remains possible.
- Check where the provider’s servers are located if data residency rules apply.