What is Llama? — a model family for beginners

An app says it uses Llama — what does that actually mean?

You see “This assistant is based on the Llama language model” in an app’s description, or assume that Meta’s own AI assistant must use Llama because Meta developed it. In April 2026, that assumption was already wrong for Meta’s own product: Meta switched the Meta AI app and meta.ai website to the new Muse Spark model, and Llama no longer powers them. The same company and product name therefore do not tell you which model an app currently uses — you need to check separately, and it is often surprisingly easy.

Llama itself is a family of language models published by Meta: a set of models in different sizes that many other companies can use in their own apps instead of Meta. This is possible because Llama is released with open weights — the model’s internal parameters can be downloaded and run on anyone’s own hardware, unlike a closed model accessible only through its maker’s service. Identifying whether your particular app uses Llama also tells you whose terms and which licence apply to it.

Find out which model is behind your app

1. 1

   Open an app you already use

   Choose an AI assistant or app you use regularly — for example, your phone’s built-in assistant, a chat tool at work or a browser-based service.

2. 2

   Look for the model’s name in settings or on the About page

   The model’s name is usually in the app’s settings, About page, help or frequently asked questions. Its exact location, and whether the information is provided at all, depend on the app, your account and your region.

3. 3

   If you cannot find it, compare two official pages

   Then open the official page of another AI provider, such as OpenAI or Google, and compare its stated model licences, weight availability and ways to use the model with Llama. A chat service may offer several models: check the specific model, rather than assuming the service name tells you whether its weights are open.

4. 4

   Note down three things

   Is the model released with open weights or is it closed? Which company provides it? And where did you check — the app’s own source or the model maker’s page? These three answers already take you a long way.

Try this

An app’s description says: “Our app uses open Llama AI technology.” Explain exactly what this sentence promises me as a user and what it does not tell me about the model, licence or privacy. Finish by listing three questions I should ask the app’s developer before trusting it with something sensitive.

Check and improve

Suppose the app’s developer answered your questions as follows: the app uses the Llama 3.1 model on its own servers in Europe, and your messages are not used for further model training. Update your earlier answer using this information: explain which of your earlier concerns it resolves and which remain unanswered.

This refinement teaches you that one good answer is not enough. The same list of questions works as a checklist whenever the app or its developer provides new information about the model.

What open weights actually mean

An app says it uses Llama and calls it an open model.

You think that because the model is open, you can use it for anything without restrictions.

Explain the intended use of a fictional Llama model, but separate information about the model, the service providing it and its licence. Finish by naming what I should check in the documentation for the particular service I use.

Once you understand the basics

Next, you can compare Llama with other model families, learn to read a complete model card — the maker’s published summary of a model’s features, limitations and licence — and watch for later model changes in the app, just as Meta changed its own assistant. To explore the idea of open models more generally, continue to the guide on open and small models.