Learn AI / Compare /Open-source vs open-weight

Open-source vs open-weight

Downloadable is not the same as open.

The short answer

An open-weight model publishes the trained numbers, so anyone can run it. Open-source in its full sense would also mean the training data, the code and a licence without use restrictions. Most models called "open" are open-weight, and the distinction matters for procurement.

Open-source AI

AI Foundations · Intermediate

In one sentence: AI systems released under licence terms that let others inspect, use, modify and redistribute source code and other materials needed to run them.
Why it matters: For non-specialists, this term matters because openness affects procurement, security review, legal risk, vendor dependence, cost and public trust. An open-source AI project may let your organisation inspect parts of the system, host it internally, commission improvements and avoid being locked into one supplier. But the label alone does not answer the most important questions: exactly what is open, under what licence, with what documentation, and with what safeguards. In policy and leadership settings, those details shape accountability and practical control.
The everyday version: It means source code and needed system materials are shared under licence rules that let others inspect, run, change and redistribute them.
The one thing to remember: Open-source is a claim about released components and licence rights, not a mark of quality, safety or approval. Check what is actually available. Weights alone make an open-weight release, not necessarily an open-source one.
The commonest misconception: If an AI model is open source, it must be safe and trustworthy. Open release can improve scrutiny because more people can inspect or test some parts, but it does not guarantee safe behaviour, legal compliance or good performance. A poorly documented or biased system can be open source too.

Read the full page

Open-weight model

AI Foundations · Intermediate

In one sentence: A model whose trained numerical parameters are publicly released so others can run, inspect or adapt it, even if the training code, data or licence are not fully open.
Why it matters: The phrase affects procurement, risk assessment and governance. If weights are released, an organisation may be able to run the model itself, test it more deeply and adapt it to local needs. But that does not tell you whether the training data are known, whether the licence permits your use, or whether the system is safe, fair or suitable for sensitive work. Knowing the distinction helps avoid overclaiming transparency.
The everyday version: An open-weight model is a model whose trained settings have been shared so other people can run it themselves.
The one thing to remember: Open weights provide the trained parameter values, not necessarily the training data, code, full development record or unrestricted permission to use and adapt the model.
The commonest misconception: If a model is open-weight, it is open source. Not necessarily. Open weights means the trained parameters are available. Open source usually concerns software code and the legal rights to use, study, modify and share it. A project can release weights while keeping code, data or some usage rights restricted.

Read the full page

Proprietary model

AI Foundations · Intermediate

In one sentence: An AI model whose owner uses licence terms or technical controls to restrict access to or reuse of key components, such as its weights or code.
Why it matters: For decision-makers, "proprietary" signals a governance choice, not just a technical one. It affects procurement, cost, bargaining power, audit access, security review, data handling, business continuity and exit options if a supplier changes terms or stops a service. It also affects accountability after harm or error. If important parts are closed, an organisation may have less ability to inspect how the system was built, test it independently, reproduce results or move the workload elsewhere. That does not make proprietary models automatically bad or unsafe, and open models are not automatically better. It means the trade-offs must be examined clearly before adoption.
The everyday version: A proprietary model is controlled through restrictive access or licence terms. Important components or rights to modify and redistribute it remain with its owner.
The one thing to remember: A proprietary model can be usable without being inspectable or reusable. Check which components and rights remain under provider control, not merely whether the service produces an answer.
The commonest misconception: A proprietary model is simply a model you have to pay for. Price and openness are different questions. Some paid services share more technical access than others, and some free services still keep weights, code and training details closed. The core issue is control over inspection, modification, hosting and redistribution, not only whether there is a fee.

Read the full page