Proprietary model

The GPT, Claude, and Gemini families are the canonical examples: access flows exclusively through the creator's API and sanctioned cloud partners, at prices the creator sets. Proprietary models have historically defined the capability frontier, though the gap to the best open-weights models has narrowed repeatedly.

The upside is turnkey access to top capability with no infrastructure burden and enterprise-grade support. The downsides are pricing power resting with the vendor, data governance depending on contractual promises rather than physical control, and deprecation risk — the vendor decides when a model version is retired.

Most organizations mitigate lock-in by writing provider-agnostic abstractions and keeping evals ready, so switching models is a configuration change plus a regression run rather than a rewrite.