Open Weights vs Open Source vs Open Platforms

Open Weights, Open Source, and Open Platforms Are Different

Open weights give people access to a trained model's parameters under a license. Open source grants the rights and materials required by an applicable open-source definition. An open platform lets outside developers or manufacturers participate under the platform owner's rules. An open standard is a shared specification that more than one implementation can follow.

These ideas can overlap, but one does not prove the others.

The short comparison

Open weights describe access to model parameters

A model's weights are the numerical parameters produced during training. When a company makes those weights downloadable, developers may be able to run the model on their own infrastructure, fine-tune it, inspect behavior, or use a hosting partner.

The license determines what they may legally do. A model can have widely available weights while restricting certain users, purposes, redistribution, or competitive uses. Access to weights also does not necessarily include the training data, complete data-processing code, or everything required to reproduce the model.

That is why open-weight is often the clearest factual label. It says what is accessible without assuming broader freedoms.

Open source is a license and freedom claim

For software, the Open Source Initiative's definition requires more than visible code. The distribution terms must allow use, modification, and redistribution without discriminating against people, groups, or fields of endeavor.

AI systems complicate the picture because weights, training data, data-processing code, model code, and documentation can have different owners and terms. The Open Source AI Definition addresses the materials and freedoms needed to use, study, modify, and share an AI system.

The safest editorial practice is to name the license and the available materials. If a vendor calls a model open source but a recognized standards body disputes that label, explain the disagreement rather than choosing the marketing term by default.

An open platform still has an owner

A platform becomes more open when it lets outside parties build products, distribute software, connect services, or manufacture compatible devices. That access can create real value. Developers can reach more users, customers can gain more choice, and partners can share infrastructure.

The platform owner may still control entry, identity, discovery, payments, technical certification, data access, and policy enforcement. Partners can be exposed to a future change in fees, APIs, store rules, or strategic direction.

Meta Horizon OS illustrates the distinction. Meta announced access for additional hardware makers and a wider path into its app ecosystem. That was an open-platform strategy. It was not the same as publishing the complete operating system under an open-source license.

An open standard makes room for independent implementations

An open standard specifies how systems should communicate or represent information. Its value comes from allowing independent implementations to interoperate without requiring every participant to use one vendor's product.

A company can operate an open platform without supporting an open standard. It can also implement an open standard inside a proprietary product. The real test is whether another organization can build a compatible implementation and whether the standard's governance and legal terms permit that independence.

Ask what freedom you actually receive

The word open is useful only when it points to a specific right or operating condition.

Before adopting a model, check the license, the materials provided, who may use them, what uses are restricted, what it costs to operate the system, and what happens when a new version arrives.

Before joining a platform, check who controls access, distribution, identity, payments, customer data, technical certification, and policy changes. Before depending on a standard, check its governance, intellectual-property terms, implementation support, and real adoption.

The right question is not whether something is open in the abstract. It is what is open, to whom, under which terms, and who can change those terms later.

Sources and editorial notes

This explainer draws on the Open Source Definition, the Open Source AI Definition, the Open Source Initiative's analysis of the Llama 3.x license, and Meta's Horizon OS partner announcement. It is a conceptual guide, not legal advice.