Review Open Weights vs. Closed Weights

Open weights and closed weights describe whether a trained model's parameter values are released for others to obtain and run or remain controlled by the developer and accessible only through approved products or services.

Open-weight models can often be hosted, inspected, modified, or fine-tuned under their license. Closed-weight models are usually accessed through a vendor interface or API without receiving the underlying parameter files.

Weight access affects deployment control, privacy options, customization, verification, cost, security, and dependency, but licenses and missing training materials can still limit practical openness.

A company compares running an open-weight model in its own environment with sending requests to a closed-weight model through a vendor API.

Open weights do not automatically mean open source, unrestricted use, transparent training data, or safer behavior.

Open weights grant access to learned parameters, not necessarily the whole recipe or every right.

* [[weights|Weights]] * [[build-vs-buy|Build vs. Buy]] * [[vendor-lock-in|Vendor Lock-In]] * [[layered-sovereignty|Layered Sovereignty]] * [[capability-provenance|Capability Provenance]]

* [Stanford CRFM, The Foundation Model Transparency Index: Open Foundation Models](https://crfm.stanford.edu/open-fms/) — research-based treatment of openness dimensions, including weights and access. * [Bommasani et al., Considerations for Governing Open Foundation Models](https://arxiv.org/abs/2403.07918) — primary research on distinctions and governance considerations for open foundation models.

* Evaluate access, licenses, infrastructure, and data separately through [[layered-sovereignty|Layered Sovereignty]].