What is Llama?

Llama is Meta's family of open-weight language models. This lesson explains what open weights mean, who Llama is for, and where it falls short.

In this lesson you will

Describe what Llama is and how it differs from a closed model like GPT or Claude
Explain the three ways you can actually use a Llama model
List the honest limits of open-weight models before you commit to one

Llama is a family of large language models published by Meta. The thing that makes it different from ChatGPT, Claude or Gemini is not the chat window. It is that Meta releases the model weights themselves, so anyone can download the model, run it on their own hardware, fine-tune it, or offer it as a service.

That single fact changes how pricing works. With a closed model there is one seller and one price list. With Llama there is no single price at all. You pay whoever is running the model for you, or you pay for the hardware and run it yourself.

Figure 1.Every LLM, open or closed, does the same job: it reads your text as tokens and predicts the next ones. Llama is one of the few families where you can see and run the whole thing.

Three ways to use Llama

Meta AI — the consumer assistant inside WhatsApp, Instagram, Messenger and at meta.ai. It runs on Meta's models and is free to use. You do not pick a model or pay per token.
Hosted API — a provider such as OpenRouter or a cloud platform runs the model and charges you per token. This is where the prices in our catalogue come from.
Self-host — you download the weights and run them on your own machine or a rented GPU. There is no per-token bill, but you pay for hardware, electricity and your own time.

Figure 2.Open weights: the same model file can be served by many hosts at many prices, or run by you with no meter at all.

Who makes it and what it is good at

Meta trains and publishes the Llama family. The current generation, Llama 4, ships in two sizes in our catalogue: Llama 4 Scout and Llama 4 Maverick. Older but still widely used models include Llama 3.3 70B Instruct and the small Llama 3.1 8B Instruct. Meta also lists newer Muse Spark and Muse Glimmer models through hosted APIs, which we cover in the next lesson.

Llama's strengths are practical rather than glamorous. It is cheap per token on hosted APIs. It can be run privately, which matters for regulated data. It can be fine-tuned on your own examples. And because many hosts serve it, you are not locked into one vendor's price list or terms of service.

Honest limits

Open-weight models generally trail the best closed models on the hardest reasoning and coding tasks. If you need frontier performance on difficult work, a Llama model may need more prompting and more checking. Quality also varies between hosts: the same model can be served at different precision, with different context limits and different speeds.

There is no official first-party consumer subscription like ChatGPT Plus. Meta AI is free, but it is a general assistant, not a developer tool with model selection and settings. If you want to choose a specific Llama model, you will be using a hosted API or running it yourself.

Knowledge check

What is the key difference between Llama and a closed model such as GPT or Claude?

Lesson FAQ

Is Llama free?

The weights are free to download under Meta's licence, and the Meta AI assistant is free to use. Hosted API access is charged per token, and self-hosting costs hardware and electricity.

Is Llama as good as ChatGPT or Claude?

For everyday tasks it is often close enough, at a fraction of the per-token price. For the hardest reasoning and coding tasks, closed frontier models usually still lead. Test on your own work before deciding.

Do I need a GPU to use Llama?

No. A hosted API such as OpenRouter runs it for you. You only need a GPU if you want to self-host, and small models such as Llama 3.1 8B can run on a decent laptop.