Lesson 24 of 28 · Using AI well

What is open-source AI?

Open-source AI is AI whose building blocks are published so anyone can use, study, modify and share them. For a model, that means the weights (the numbers it learned), the code, and enough information about the training data to rebuild it. Many "open" models release only the weights under a licence with conditions; these are more precisely called open-weight models.

What "open" means for an AI model

In ordinary software, open source means the source code is public under a licence that lets anyone use, change and share it. An AI model has more parts: the code that defines and trains it, the training data, and the weights, the millions or billions of numbers adjusted during training that hold what the model learned.

The Open Source Initiative's Open Source AI Definition 1.0, published in October 2024, asks for all the parts needed to understand and change a system: the weights, the code and detailed information about the data. In practice, releases fall on a spectrum:

Closed: you can use the model only through the vendor's app or API.
Open-weight: the weights can be downloaded and run, often with licence conditions, but training data and code may stay private.
Open source: weights, code and data information are all available under terms that allow use, study, change and sharing.

Why people choose open models

A kitchen analogy helps. A closed model is a restaurant dish you can order but not take apart. An open-weight model is a finished cake you can take home and decorate. Open source gives you the cake plus the recipe and the list of ingredients. More openness means more control.

That control matters in practice. Organisations run open models on their own servers so sensitive data never leaves their network, fine-tune them (train them further on their own examples) for a specialist job, avoid depending on one vendor and keep costs predictable at high volume. Researchers can inspect how a model behaves, and individuals can run small models offline on a laptop.

The trade-offs

Free to download is not free to run. Large models need powerful graphics processors (GPUs) and plenty of memory, or a paid hosting provider. Whoever runs the model also takes on the security, updates and safety filters that a closed-model vendor would otherwise handle.

Licences differ widely. Some allow almost any use; others restrict commercial use, very large companies or certain applications. Meta's Llama models, for example, come with Meta's own licence and conditions, while DeepSeek has released the weights of several models under a permissive licence. Always read the licence for the exact model and version before you build on it.

Where you meet it in real tools

You may already use open models without knowing it. Many apps and AI features are built on an open-weight model behind the scenes, and cloud platforms let developers call open models through an API just like closed ones.

If you are curious, desktop apps can download a small open model and run it entirely on your own computer, with no internet connection needed once it is installed. It will usually be slower and less capable than large hosted models, but it shows clearly what "running the weights yourself" means.

Example

Run a small open model in a few lines

Code sample

# pip install transformers torch
from transformers import pipeline

# Replace MODEL-ID with a small open model whose licence allows your use
generator = pipeline("text-generation", model="MODEL-ID")

result = generator("Explain open-source AI in one sentence:", max_new_tokens=40)
print(result[0]["generated_text"])

This Python sample uses the open-source Hugging Face Transformers library. It downloads the model's weights once and then runs them on your own machine, which is exactly what open weights make possible. Start with a small model, and read its licence first.

Key points

Open-source AI publishes the parts needed to use, study, modify and share a system.
Many "open" models are really open-weight: downloadable weights, but not all the data or code.
Open models give control over data, customisation and cost, but you carry the hosting and safety work.
Licences vary; read the one for the exact model before commercial use.
Free to download is not free to run: large models need serious hardware.

Try it on Mindber

Find open-source LLMs

DeepSeek on Mindber

Compare open and closed models

Glossary: open-weight model

Frequently asked questions

Is open-source AI free?

The licence may cost nothing, but running the model does not. You pay for hardware, electricity or a hosting provider, plus the time to set up, secure and update it. For light use, a paid closed service can work out cheaper; at high volume or under strict data rules, running an open model yourself can pay off.

What is the difference between open-source and open-weight AI?

An open-weight model shares its trained weights so you can run and often fine-tune it, but the training data and code may remain private and the licence may restrict use. Open-source AI, under the Open Source Initiative's definition, also requires the code and detailed information about the data, under terms that allow use, study, modification and sharing.

Is open-source AI safe to use?

It can be, with care. Download models only from the publisher's official pages, keep the software up to date and test the model on your own tasks. Because anyone can modify open weights, safety filters may be weaker or removed in modified copies, so the person or company deploying the model is responsible for adding safeguards.