Free and open AI models.

What "free" and "open" mean

An AI model is, at its core, a large file of numbers called weights, learned during training, plus the code that runs them. When people talk about a "free" or "open" model, they usually mean that the weights can be downloaded and run by anyone.

That is not the same as open source in the traditional software sense. It helps to separate three things:

• Open weights: the trained model can be downloaded. The training data and training code may not be published.
• Open source AI: a stricter idea. The Open Source Initiative published a definition in 2024 that expects, as well as the weights, enough information about the training data and the code used to train and run the model for someone else to study and rebuild it.
• Free of charge: you pay nothing to download. It does not mean you may use the model for any purpose. That depends on its licence.

Many of the most popular downloadable models are open-weight rather than fully open source. Being clear about which one you are dealing with avoids wrong assumptions about what you are allowed to do.

Licences decide what you may do

Every downloadable model comes with a licence, and the licence, not the price, decides how you may use it. Broadly, you will meet three kinds:

• Permissive open-source licences, such as Apache 2.0 and MIT. They allow commercial use, modification and redistribution, provided you keep the copyright and licence notices. Apache 2.0 also asks you to state which files you changed.
• Custom model licences written by the model's publisher. They often allow commercial use but add conditions, such as an acceptable-use policy, attribution requirements or a separate licence for very large services. Meta's Llama community licences are a well-known example.
• Non-commercial or research-only licences, which forbid commercial use altogether.

Three habits keep you safe. Read the licence of the exact model and version you use, because terms can change between versions. Keep the licence and notices with any copy you distribute. And do not suggest that the original publisher endorses your product unless they do.

Licences cover the model, not the data you put into it or the laws that apply to your use. Data protection and sector rules still apply to what you do with it.

Choosing a model

Open models are shared through model hubs, the largest being Hugging Face, and through the publishers' own sites. Each well-published model comes with a model card: a description of what the model is for, how it was trained, what it should not be used for, its known limits and its licence. Read it first.

The main things to compare are:

• Size, measured in parameters (for example 0.6 billion or 7 billion). Larger models are usually more capable but need more memory and are slower.
• Context length: how much text the model can take into account at once.
• Languages and tasks it was trained and tested for.
• Licence, as in the previous lesson.

Published benchmark scores are a useful starting point, but they rarely match your exact task. The reliable test is a small set of your own real examples, with the answers you expect, run against each candidate model.

Also check the file format. Some older formats can run arbitrary code when the file is loaded, so a model from an unknown source is a security risk. Formats designed to store only weights, such as safetensors, avoid that problem.

Running a model yourself

The main reason to use an open model is that you can run it on hardware you control, so your data never has to leave it. The first question is memory. A rough guide is the number of parameters multiplied by the bytes stored for each one:

• at 16-bit precision, each parameter takes 2 bytes, so a 7-billion-parameter model needs roughly 14 GB for its weights alone;
• at 4-bit precision, it needs roughly a quarter of that, around 3.5 GB, plus working memory.

Reducing the precision like this is called quantisation. It makes models small enough for ordinary laptops, and even phones for the smallest models, at the cost of some quality. How much quality is lost depends on the model and the task, so test it.

Common tools include llama.cpp, which runs quantised models efficiently on ordinary computers, Ollama, which packages models for easy local use, and vLLM, which serves models on GPU servers for many users at once. A graphics processor (GPU) speeds things up considerably, but small models can run on a normal processor.

Running a model yourself also makes you responsible for it: keeping the software updated, controlling who can reach it, and recording which model version is in use.

Using open models responsibly

Open models share the weaknesses of all language models, and some of the responsibility shifts to you:

• They can be confidently wrong. Models produce plausible text, not verified facts. Check anything that matters.
• Their knowledge stops at a training date and may be missing local context, languages or recent events.
• They can reflect bias in their training data. Test them on the people and cases they will actually affect.
• They come without a warranty. Open licences typically provide the model "as is".
• They can be manipulated by instructions hidden in the text they are given to read (prompt injection).

Good practice is to pin the exact model version, keep a record of its licence, test it on your own cases before and after any change, and keep a person, or fixed rules, responsible for any decision that affects people or money.

Saolix Eagle runs its AI features on locally hosted open models, and nothing is sent to an external AI service. The AI only advises; approved rules and people make the decision. Sōli AI's first model, Soli Lite, is being built on Qwen3-0.6B, an open model released under the Apache 2.0 licence, and is still in development.

Your certificate of completion

Enter your name as you want it to appear, then save the certificate as a PDF. In the print window, choose Save as PDF.

Saolix does not record who takes this course, so it cannot verify these certificates. The certificate confirms completion of a free self-paced course and is not an accredited qualification.