GLOSSARY
Open Weights (Open-Source Models)
Models whose trained parameters are publicly downloadable, so anyone can run, audit or adapt them on their own hardware.
Open-weights models publish their trained parameters for anyone to download — unlike closed models such as GPT-4 that are only reachable through a company's API. The result is an ecosystem: DeepSeek, Qwen, Flux and Stable Diffusion can be self-hosted, fine-tuned, audited and combined without asking permission.
'Open weights' is not the same as full open source (training data and code often stay private), but the practical freedoms are real: data never leaves your servers, per-token costs disappear, and the community collectively improves on each release. Hugging Face is the marketplace where this ecosystem lives.