What is a Closed Model?

Closed models — such as those offered by Anthropic, OpenAI, and Google — contrast with open-weight models where the parameters can be downloaded and run independently. With a closed model, developers send requests to a hosted API and receive completions back; they cannot inspect the weights, fine-tune the base architecture freely, or run the model on their own infrastructure without the provider's involvement. The tradeoff is real: closed models tend to lead on raw capability and come with a managed service — uptime guarantees, safety tuning, and regular improvements without any local maintenance burden. Open-weight models trade some capability for control: they can be self-hosted, audited, fine-tuned deeply, and run without sending data to a third party, which matters for certain compliance or latency-sensitive use cases. Many production systems end up using a closed model as the default reasoning engine while keeping an open-weight or alternate model in the routing logic as a fallback or for cost-sensitive tasks, rather than picking one exclusively.