Estimated reading time: 6 minutes

Key points:

  • With ChatGPT, Claude and Copilot your input goes to the servers of a US provider. With open source models you can avoid that.
  • You run an open model in your own environment or with a European provider, so your data stays in-house or within Europe.
  • That mainly matters for sensitive data: personal data, medical or legal information, or data you would rather not share for other reasons.
  • The price you pay is more setup and maintenance, and sometimes a model that is just a little weaker than the very best closed alternative.
  • It is not an all-or-nothing choice. Many organisations use closed models for ordinary work and an open model for the sensitive parts.

Contents

  1. What is open source AI?
  2. Why does it matter where your data goes?
  3. When is open source the better choice?
  4. What does open source AI cost?
  5. Open or closed: how do you choose?
  6. How we can help
  7. Frequently asked questions
  8. Sources

With the well-known AI tools, you send your question to the servers of OpenAI, Anthropic or Microsoft. For a lot of work, that is fine. For sensitive data it grates: you don't always know where the data sits, and you hand over a measure of control. Open source AI offers an alternative, and in this conversation it is often the overlooked option.

Below you will read what open source AI is, when it is the better choice, and what the trade-off is.

What is open source AI?

Open source AI is about models that their makers release themselves, so you can run them wherever you want: on your own servers, or with a provider you choose. Well-known examples are the models from the European company Mistral and the Llama models from Meta.

The difference with ChatGPT, Claude and Copilot lies in where the model runs. Those are closed: you use them through the provider's servers, and you have no say over where that happens. An open model runs in an environment you control. That shifts control over your data from the provider to you.

Why does it matter where your data goes?

For a draft email it does not matter where your data goes. For a file with personal data, a medical report or a confidential contract, it does.

With a US provider you cannot always guarantee that data stays within Europe, even when the provider says it complies with the GDPR. For some organisations that is an inconvenience, for others a hard limit, for example in healthcare, government or the financial sector.

With an open model that you run in-house or within Europe, your data does not go to a server outside your control. You decide where it sits, who can reach it, and how long it is kept. For anyone working with sensitive data, that is the most important argument.

Wireframe iceberg with a small visible tip above the waterline and a much larger hidden body below, illustrating the visible and hidden costs of open source AI

When is open source the better choice?

Open source is not the right choice for everyone, but in a few situations it clearly is:

  • You work with sensitive data. Personal data, medical or legal information, or data that for other reasons may not leave your premises.

  • You face strict requirements. Rules or agreements that prescribe where your data may be stored.

  • You use AI at scale. With heavy use, the cost of closed models can add up. Your own model can then be cheaper over time.

  • You want to be independent. No reliance on a single provider that can change its prices or terms.

If none of this applies to you, and you mostly want to get started quickly without sensitive data, then a closed tool is usually the easier route.

What does open source AI cost?

Open source is not free in the sense of effortless. There is a price for the control you get back.

You need an environment where the model runs, and someone to set it up and maintain it. That takes knowledge and attention you don't need with a closed tool, where you log in and it works. The cost shifts from a subscription per user to infrastructure and management.

On top of that, the very best closed models are sometimes still a step ahead on the hardest tasks. For a lot of work you won't notice that difference, but it is fair to mention. So the choice is a trade-off: how much is control over your data worth to you, set against the convenience and the sometimes slightly higher quality of a closed tool.

Wireframe balance scale weighing a small object against a wireframe house, illustrating the trade-off between a closed AI tool and keeping data in-house

Open or closed: how do you choose?

You don't have to pick one of the two. The practical approach is often a combination: closed models for ordinary work where the data is not sensitive, and an open model for the parts where your data has to stay in-house.

Which split fits you depends on your data, your requirements and your scale. That is the same trade-off as the choice between ChatGPT, Claude and Copilot, which we describe in ChatGPT, Claude or Copilot: which fits which work. Open source is the fourth option there, and for organisations with sensitive data often the serious one.

How we can help

The right choice depends on your data and your situation, not on which model is most popular. We map that out: which tasks can safely stay with a closed tool, and which belong on an open model under your own management. We compare the options on quality, cost and where the data ends up, and help set it up within your own environment.

If you want to know whether open source is right for your organisation, and where the line then lies between what may go out and what stays in, we look at that together in AI in Business.

Frequently asked questions

Is open source AI free?

The model itself is free to use, but you need an environment to run it and someone to manage it. The cost shifts from a subscription to infrastructure and maintenance.

Is open source AI less good than ChatGPT or Claude?

For a lot of work you won't notice a difference. On the hardest tasks the best closed models are sometimes still a step ahead. The advantage of open source is control over your data, not the highest score.

Does my data really stay within Europe with open source?

It can, if you run the model in your own environment or with a European provider. You decide where it sits. With closed US tools you don't have that guarantee.

Which open models are there?

Well-known examples are the models from the European company Mistral and the Llama models from Meta. Which model fits depends on your task and your requirements.

Do we then have to switch entirely?

No. Most organisations combine: closed tools for ordinary work, an open model for the sensitive parts. You choose per type of data, not all at once.

Sources