Open Source

Every program is written by people in a programming language. This written text is the source code, in other words the blueprint of the software. Normally, users only get the finished program, not this blueprint. Open source means that the blueprint is published. Anyone who wants to may read it, change it, and pass on their own version of it. A cookbook is a fitting comparison: you don’t just get the dish, you get the recipe for it too.

What openness changes for users and companies

Open code can be checked. With a messaging app, this lets you verify whether messages are really encrypted. With closed software, you simply have to trust the manufacturer. Security researchers also find bugs in open code that the team itself missed.

For companies and authorities, it’s also about independence. Anyone using closed software is dependent on a single provider. If that provider raises prices or discontinues the product, you have a problem. Experts call this lock-in, meaning being locked into one provider. With open-source software, you can maintain the code yourself if necessary, or pay another company to do so.

A common misconception: open source doesn’t automatically mean free of charge or usable however you like. Some licenses require that your own modifications must also be open. Others allow almost anything, including incorporation into commercial products. The difference is legally important and often overlooked.

Licenses, repositories, and who is allowed to contribute

Technically, the code usually lives in a public repository, an online storage space for projects. The best-known platform for this is GitHub. There you can see every change along with date and author. Anyone wanting to improve something submits a change proposal, which the core team can accept or reject.

The license is decisive—a short legal text in the project folder. It sets out what is allowed. The MIT license permits practically any use, including in paid products. The GPL, by contrast, requires that derivative programs also be published openly. Without a license, code is not legally free to use, even if it’s visible online.

With AI models, the situation is more complicated. A language model doesn’t just consist of code—above all, it consists of weights. These are billions of numbers that emerged during training. Many providers publish these weights for download but keep the training data secret. You can then use and adapt the model, but you can’t trace what it learned from. Experts therefore speak more precisely of open weights rather than open source.

From Linux to Llama and DeepSeek

Open-source software runs everywhere, even if you barely notice it. The Linux operating system powers the majority of all web servers. Android is built on it. The Firefox browser is open, as are large parts of Chrome. The programs researchers use to develop AI are also mostly open source.

In business news, the topic usually revolves around AI models. Meta releases its Llama models with open weights, as does the Chinese company DeepSeek. OpenAI and Google, by contrast, keep their strongest models closed and only sell access to them. Behind this lies a strategic question: openness brings adoption and developers, while closedness brings direct revenue.

For users, this has practical consequences. Open models can be run on your own hardware without sending data to a provider. Hospitals or law firms make use of this because they aren’t allowed to share sensitive information. If you read in a report that a model is open source, it’s worth asking exactly what is open: the code, the weights, or also the training data.