# Open vs. Closed Models

**Track:** Model Foundations — AI for Entrepreneurship — model foundations (5)
**Framework / surface:** venture strategy
**Level:** Intermediate
**Prerequisites:** Fine-Tuning Intuition

**In one line:** Closed rents frontier quality by the token; open weights buy control — local generation, fine-tuning rights, provenance, and a cost floor you own.

## Theory, aesthetics & inspiration

Closed models — GPT, Claude, Gemini — are reachable only through an API: capability rented by the token, weights kept home. Open-weight models publish the parameters themselves: Stability AI's public release of Stable Diffusion in August 2022 ignited an ecosystem of fine-tunes and tools almost overnight, and Meta's Llama family ships weights under a community license with its own restrictions — the Open Source Initiative's 2024 Open Source AI Definition reserves the term for far fewer models than the label gets applied to. The trade for a creative company: closed buys frontier quality with zero operations; open buys control — local generation, fine-tuning rights, auditable provenance, a cost floor you own, and independence from a vendor's taste and terms of service. Durable studios usually run both, and know exactly which capabilities sit on which side.

**Founder question:** Which capability in your stack must you own outright, and which are you happy to rent?