Llama models

Llama is the family of large language models that Meta has released since 2023. What sets it apart is that the model weights are published: anyone can download them, run them on their own hardware or build products on them under Meta's license. These pages explain each model in plain language, so you can tell what it is good at and when to use it. If you are new to the topic, start with what Llama is.

Models you can use in our chat

Four Llama models answer in the Llama.Online chat, served by a cloud inference provider:

• Llama 4 Maverick, the default: a mixture-of-experts model from April 2025 that reads text and images.
• Llama 4 Scout: also reads images and accepts the longest inputs, useful for big documents.
• Llama 3.3 70B: a text-only all-rounder from December 2024.
• Llama 3.1 8B: small and quick, for simple tasks.

Other Llama models

Some pages cover models that are not in our chat but matter for anyone using Llama: Llama 3.2, with small text models for phones and vision models, and Llama Guard 4, a safety model that checks prompts and answers. The Llama history timeline traces every release from the first Llama to Llama 4.

Llama 4 Maverick and Scout are still the newest models to carry the Llama name. In 2026 Meta began releasing new models under a different brand, Muse, including Muse Spark and the open-weight Muse Glimmer and Muse Spark Open. They are separate from the Llama family and are not available in our chat.

Comparisons and practical guides

Wondering how Llama stacks up? Read Llama vs ChatGPT, Llama vs DeepSeek or Llama 4 Maverick vs Scout. To use Llama outside our site, see how to run Llama locally, how to call it through an API, and what the Llama license allows.