Open Access Terminology and Rights Matrix

Open Access Terminology and Rights Matrix

Open weights, open source, open platforms, and open standards answer different questions.

Open weights describes access to learned parameters. It does not, by itself, establish access to training data, data-processing code, training code, optimizer state, evaluation artifacts, or unrestricted legal rights.

Open source is a standards claim about usable, inspectable, modifiable, and shareable material under qualifying terms. The Open Source AI Definition 1.0 extends that inquiry to the preferred form for modifying an AI system, including data information, code, and parameters. It is a standards position, not a court judgment.

An open platform lets approved participants build, distribute, or interoperate through a platform owner’s interfaces and rules. Access can still depend on accounts, stores, certification, technical requirements, identity systems, commercial terms, and roadmap decisions.

An open standard is a shared technical specification with an identified governance and adoption path. A product can implement an open standard while remaining proprietary. A platform can publish an interface without surrendering control over the service behind it.

The Meta Llama 3 Community License grants substantial use, modification, and redistribution rights but also imposes conditions, an acceptable-use policy, a restriction on improving another large language model, and additional commercial terms for very large services. The Open Source Initiative’s Llama analysis concludes that the Llama 3.x license does not satisfy its open-source standards.

Meta’s Horizon OS announcement describes third-party hardware access and a wider app ecosystem. It does not announce a complete source-code release under an open-source license.

Public language should therefore name the artifact, rights, restrictions, dependencies, and date. “Open” alone is not enough for a technical, procurement, or legal decision.