Llama 4 Maverick
Newest Llama in our chat, reads images
Llama 4 Maverick and Scout, free in your browser
Talk to Meta's open-weight Llama models without an account. Ask questions, write, code, translate, or show Llama 4 a photo and read the answer as it streams in.
Free daily messages · no sign-up · works on phones and computers
Four Meta Llama models are in the chat. Llama 4 Maverick is the default, and you can switch from the model menu at any time.
Newest Llama in our chat, reads images
Reads images, handles the longest texts
Strong all-rounder for text, no images
Smallest and quickest, for simple tasks
A clean, quick chat around real Llama models, with the extras that make it useful every day.
Every answer comes from one of four open-weight Llama models released by Meta. You choose which one in the model menu.
Attach or paste a photo, screenshot or chart and ask Llama 4 Maverick or Llama 4 Scout what it shows.
Writer, summarizer, translator, paraphraser, math solver and more, each with a simple form and a result you can copy.
Sixteen built-in assistants and a library of prompt templates give you a strong start on common tasks.
Chats stay in your browser with search, pins, export and import. Share a read-only copy with a single link.
Use the site in 26 languages and write to Llama in your own. Switch between light and dark themes any time.
Pick a tool, fill in one box and get a result. Every tool page also explains how to get better output.
Start a chat with a tutor, an editor, a travel planner or a mock interviewer. Each assistant is a set of instructions you can read and adapt.
Write in any language, add a photo when you use Llama 4, or start from a prompt template or an assistant.
Keep Maverick for most tasks, pick Scout for very long texts, or Llama 3.1 8B when you want the quickest reply.
Ask follow-up questions, regenerate or edit your message, then copy, export or share the result.
Meta released the first Llama in February 2023 and Llama 4 in April 2025. Its later models carry a new name, Muse. Follow the whole story on our timeline.
Side-by-side looks at Llama versus ChatGPT, DeepSeek, Qwen, Mistral and Gemini, and between Llama versions.
Practical ideas for students, teachers, developers, writers, marketers and small business owners, with prompts you can open in one click.






Llama weights can be downloaded. Tools like Ollama, LM Studio and llama.cpp run the smaller models offline, if your hardware is up to it.
Run Llama locallyCall Llama models through an OpenAI-compatible API, with code samples in curl, Python and JavaScript and current prices.
Llama APIShort, practical guides: writing prompts, checking AI answers, choosing a Llama model, privacy and more.






Llama is a family of large language models made by Meta. Unlike most chatbots, its model weights are published, so anyone can download them, run them or build products on them under Meta's license. Llama.Online lets you chat with four of these models in your browser.
Yes. You pay nothing and need no account; the site is supported by ads shown around the content, never inside answers. Each visitor gets a daily allowance of free messages, currently 100 per day, which resets every day.
Llama 4 Maverick (the default), Llama 4 Scout, Llama 3.3 70B and Llama 3.1 8B. All four are official Meta releases run by a cloud inference provider, and you can switch between them at any time in the model menu.
No. Llama.Online is an independent website that is not run or endorsed by Meta. Meta AI is Meta's own assistant inside its apps; here you talk to Meta's open-weight Llama models through our free chat.
The newest Llama models are Llama 4 Maverick and Llama 4 Scout, released in April 2025, and both are in our chat. Meta's later models use a different name, Muse, and are not available here.
Yes, with Llama 4 Maverick and Llama 4 Scout: attach, paste or drop one image per message and ask about it. Llama 3.3 70B and Llama 3.1 8B read text only, and none of the models can create images.
No. The chat cannot browse the internet, and the models' knowledge stops at a cut-off date: August 2024 for Llama 4 and December 2023 for Llama 3.3 and Llama 3.1. Check anything recent or important with a reliable source.
Your conversations are saved only in your own browser, not in an account on our server. Messages pass through our server to the AI provider to produce an answer, and we do not store their text.
You can write in any language and Llama will usually reply in it. Meta officially lists 12 supported languages for Llama 4, including English, Spanish, Portuguese, Hindi, Indonesian, Arabic, Thai, Vietnamese and Tagalog, so quality can be lower in others.
Yes. Meta publishes the weights, and tools such as Ollama, LM Studio and llama.cpp can run the smaller Llama models offline. Bigger models need far more memory, so read our run-locally guide before you download.

Llama.Online is a free Llama AI chat that runs in your browser. Behind it are four open-weight models released by Meta: Llama 4 Maverick and Llama 4 Scout, which also understand images, plus Llama 3.3 70B and the lightweight Llama 3.1 8B. We do not host the models ourselves. A cloud inference provider runs them, and our site adds what makes a chat practical: saved history, assistants, a prompt library, twelve AI tools and an interface in 26 languages.
We are a small independent team, not Meta, and the site is paid for by ads placed around the content. There is no account and no subscription, just a daily allowance of free messages. Your conversations stay in your browser unless you decide to share one. Llama can be wrong, so treat its answers as a useful draft and check the facts that matter to you.
No account and nothing to install. Open the chat, type a message and see what Llama can do.
Open Llama chat