The world of artificial intelligence is constantly and rapidly changing – new models, tools and announcements are made almost daily. A new trend in the AI world stands out here: Making AI models publicly accessible.

But what exactly does that mean? And how open does “open” actually mean?

OpenAI recently published a so-called “open model” – GPT-4-turbo. This raises many questions. Is it really open source? Can developers do whatever they want with it? Or is it more of a strategic decision with clear boundaries?

This blog post takes a look at what exactly an “open model” is, how it differs from traditional open source models and what opportunities and challenges it presents.

Because one thing is clear: dealing with openness in AI has long been more than just a technical detail. It is about trust, control and innovation.

What does “Open Model” mean?

The term “open model” is currently causing a stir – but what does it actually mean? Put simply, an “open model” is an AI model that is publicly accessible. However, this does not automatically mean that it is also completely open in the sense of “open source”. The difference lies in the rights of use. While open source models are not only freely available, but can also be freely modified and redistributed, restrictions often apply to open models.

A current example is OpenAI with the publication of GPT-4-turbo. The model itself can be used publicly, for example via the API or in ChatGPT, but the underlying code and training data are not openly accessible or customizable. It is therefore an open model, but not an open source model.

This contrasts with projects such as Meta’s Llama 3 or Mistral. These models are published with open licenses, often including code and model weights. Developers can use them to train or adapt their own versions.

In short: “open model” means accessible, but not necessarily free to use. If you want to work with AI, you should therefore take a close look at the rights and options that come with it.

Why is OpenAI taking this step?

Why is OpenAI opening up with an “Open Model” – at least partially? There are probably several reasons for this:

• One of them is quite clear: the growing pressure from open source AI models such as Llama 3 or Mistral. These models are made publicly available, often with very few restrictions. This gives developers a lot of freedom – and puts pressure on companies like OpenAI to remain competitive.
• There has been increasing criticism from the developer community recently: many are calling for more transparency and openness, especially when AI systems are becoming so central to the future. OpenAI has responded to this – albeit within a framework that continues to set certain limits.
• Another possible motivation is control. With its own open model, OpenAI can better shape how “open AI” is used – without completely losing control, as could happen with true open source models. In this way, regulation can be actively shaped.

Officially, OpenAI says the move is part of its mission to make AI widely available, but responsibly. Open, but not completely free. A balancing act between innovation, security and competitiveness.

Chance vs. Herausforderung

An open AI model like that of OpenAI opens up many opportunities:

• Openness increases transparency. Users and developers understand how the model works and build trust.
• This promotes AI innovations. More people work with the model, allowing new ideas and applications to be developed more quickly. This offers start-ups or small development teams the opportunity to keep up with large competitors.

But such a function also has its disadvantages:

• Security is crucial here: with open models, there is an increased risk that it will be misused, for example to spread disinformation or for automated attacks.
• The license conditions must also be observed. Not every “open” model can really be used or modified freely.
• Openness also means that others can build on this technology and perhaps even develop competing products. The balancing act between open access and commercial interest therefore remains a real challenge.

What are OpenAI’s competitors?

The major AI providers pursue different strategies when it comes to openness. While OpenAI offers an “open model” with GPT-4-turbo, the actual source code and training remain behind closed doors. It is accessible, but not completely open in the classic open source sense.

• Meta operates in a completely different way. With Llama 3, the company is clearly committed to open source. The code, model weights and often sample applications are freely available – albeit under a license that restricts certain commercial uses. Nevertheless, developers can download, test, adapt and redistribute the model.
• Google, on the other hand, is rather cautious. Models such as Gemini are currently not openly accessible. They are integrated via proprietary products or APIs, but not released directly. The focus here is clearly on control.
• Another player is Mistral: the company has opted for open source models from the outset and regularly releases high-performance models with minimal restrictions.

These different approaches show: Openness has long been a competitive factor. Those who offer open models strengthen innovation and community – but also take greater risks at the same time. The decision as to how “open” a model should be is therefore also a strategic decision.

Conclusion: Can OpenAI keep up with the competition?

With the publication of GPT-4-turbo as an “Open Model”, OpenAI has taken an important but deliberately cautious step. The decision shows that there is a real need for more transparency and access – from the developer community, from companies and from society as a whole. At the same time, it is clear that OpenAI does not want to be completely open, as is the case with traditional open source projects.

The comparison with providers such as Meta, Google or Mistral makes it clear that openness is now also a strategic tool in competition. Those who are open promote innovation – but also take more risks. Those who act more cautiously protect their control and business model – but run the risk of losing touch with the community.

The “open model” is therefore full of opportunities and challenges. However, it is clear that the future of AI will not be determined by technical advances alone, but also by the question of how open and accessible this technology should be – and for whom.