Language Models at Karl AI

ChatGPT is certainly very familiar to you. Behind this chat application is the (still) best-known language model, GPT-5. You’re probably also aware that there are competitors offering their own language models—for example, Anthropic with Claude, Google with Gemini, or Mistral AI with its model of the same name.

With Karl AI, you have the key advantage of being able to choose between all of these language models without having to take out a separate subscription each time. In this article, we’ll take a closer look at what you can do with these language models in Karl AI.

What do generative language models do in Karl AI?

With Karl AI, you design prompts for interpreting empirical materials such as interviews, group discussions, observation protocols, and similar data. The prompts you create (or those provided by Karl AI) are sent—together with the empirical materials and, if applicable, additional information—via an API connection to a language model you have selected in advance, for example a Claude model from Anthropic. The model processes the input, and its interpretation is then returned to the Karl AI interface. There, you can ask follow-up questions, refine the output for your purposes, and store everything in a folder system tailored to your needs for further use later on.

Why different language models?

Language models are developed by different companies that set different priorities—for example, in the selection of training data or in how the models are trained. They differ in aspects such as the number of parameters (larger is not necessarily always better) and the algorithms used. This leads to differences in their outputs. However, results depend not only on the models themselves but also significantly on the input data—such as the prompts you provide or other contextual information given to the model.

Which model should you use in Karl AI?

There is no one-size-fits-all answer. Until about a year ago, OpenAI’s models were the most powerful and widely used. In the meantime, Anthropic’s models have caught up—and, according to some, even surpassed them. The European provider Mistral lags slightly behind but offers strong data protection, as data is processed within the EU.

Outlook

We currently have several models from the three providers mentioned above integrated into our AI interpreter. In the future, we plan to add more models from other providers. For example, we are planning to implement models from Google (Gemini) and are in contact with Apertus, a Swiss provider that is particularly interesting from a data protection and transparency perspective. Another area of focus is so-called local models—those that run on users’ own computers rather than sending data to external cloud providers. We have experimented with several local models on our server, but so far they are not yet powerful enough for us to make them broadly available.

How to get started

Theory only gets you so far. The best way to get a feel for the capabilities of these models—and the differences between them—is to try Karl AI for yourself. Take a transcript and have all models interpret it using the same prompt. You can also use our GDPR-compliant sample data for this. By comparing the results, you’ll quickly develop a sense of the similarities and differences between the models.