Llama 4 Maverick: Features, Limits and Free Chat

What Llama 4 Maverick is and who makes it

Llama 4 Maverick is a large language model from Meta, released on April 5, 2025 together with its smaller sibling, Llama 4 Scout. It belongs to Llama 4, which as of October 2026 is still the newest generation of Meta's Llama family. Like other Llama models, its weights are published for anyone to download under the Llama 4 Community License.

Maverick is natively multimodal: it reads text and images in the same conversation and answers in text. Meta built it as a mixture of experts. According to Meta's model card, it holds 400 billion parameters spread over 128 experts but activates only 17 billion of them for each token. In practice that means a lot of stored knowledge with fairly quick answers. In its launch post Meta called Maverick its "product workhorse" for general assistant and chat use, and singled out precise image understanding and creative writing.

Maverick is the default model in the Llama.Online chat. We are an independent site, not run by Meta; the model runs on a cloud inference provider and we pass your messages to it.

Where it sits in the Llama family

Llama 4 has two released models, Scout and Maverick. Both use 17 billion active parameters per token, both accept images, and both share the same knowledge cut-off of August 2024. Maverick is the bigger one, with eight times as many experts, which generally helps with nuanced writing, harder questions and detailed image reading. Scout is lighter and has a far longer context window. Older Llama models such as Llama 3.3 70B and Llama 3.1 8B are text-only. Meta also previewed a much larger Llama 4 Behemoth in 2025, but it has never been released. The full timeline is on our Llama history page.

What Maverick is good at

Everyday assistant work. Clear explanations, quick research summaries from text you paste, planning, comparing options and brainstorming.
Writing. Emails, posts, product descriptions, stories and rewrites in a different tone. Meta highlights creative writing as a strength.
Images. The model card lists visual recognition, image reasoning, captioning and answering general questions about an image. Try a photo of a menu, a chart, a handwritten note or a screenshot of an error message.
Many languages. Twelve languages are officially supported: Arabic, English, French, German, Hindi, Indonesian, Italian, Portuguese, Spanish, Tagalog, Thai and Vietnamese. Meta says pre-training covered about 200 languages, so others often work too, with more mistakes.
Coding help. Explaining code, writing small functions, spotting bugs and suggesting fixes.

How to use Maverick in our chat

Open the chat. Maverick is selected by default, or use the link below.
Type your question. To ask about a picture, attach one image per message with the attach button, by pasting or by dragging it in.
Read the answer, then ask follow-ups. You can regenerate an answer or edit your message and resend it.

Chat with Llama 4 Maverick

The chat is free with a daily allowance of messages, currently 100 per day. If Maverick is busy, you can switch to Scout or Llama 3.3 70B from the model menu.

Limits to know about

Knowledge cut-off. Maverick's training data ends in August 2024. It does not know later events, and in our chat it cannot search the web.
Confident mistakes. It can invent facts, quotes, figures or references. Check anything that matters against a reliable source; our guide to fact-checking AI answers shows how.
Image reading is not perfect. Small print, blurry photos and complex diagrams can be misread. It describes and reasons about images but cannot create or edit them.
Long inputs. Our provider offers Maverick a context window of 524,288 tokens, and Meta's card lists 1M. Our free chat, however, trims long conversations to the most recent part before sending them, so for whole books or large codebases use the API or run it yourself.
No longer the newest. Newer models from other companies, and Meta's own Muse models, are generally stronger at demanding coding and multi-step agent tasks. None of the Muse models are available in our chat.

Getting Maverick officially

Meta publishes the Maverick weights on Hugging Face as meta-llama/Llama-4-Maverick-17B-128E-Instruct; you need to accept the license on the model page before the files unlock. The license allows commercial use, but products built with it must show "Built with Llama", and companies above 700 million monthly active users need a separate agreement with Meta. For the multimodal Llama 4 models, the acceptable use policy does not grant those rights to people or companies based in the EU, while end users of products built with them are not affected. Our license page explains this in plain words.

Maverick is a big model to host yourself; Ollama lists its download at 245 GB. Most people reach it through a cloud provider instead. OpenRouter lists it at $0.1875 per million input tokens and $0.6525 per million output tokens (OpenRouter list price, Oct 2026). Meta's own hosted Llama API was shut down in July 2026.

Privacy notes

In our chat, your messages and any image you attach are sent through our server and our AI gateway to the inference provider to generate the answer. We do not store your message text, and attached images are used only for that answer. Your chat history stays in your browser. Avoid pasting passwords, ID numbers or other sensitive personal data into any AI chat. If you need full control, running Llama on your own hardware keeps everything on your machine.

How Maverick compares

Against Llama 4 Scout, Maverick usually gives richer answers while Scout handles longer inputs and costs less; see Maverick vs Scout. Against Llama 3.3 70B, Maverick adds images, more languages and newer knowledge, and Meta says it offers higher quality at a lower price; see Llama 3.3 vs Llama 4. Against closed assistants such as ChatGPT or Gemini, Maverick is older and has fewer built-in features, but you can download it and run it anywhere; see Llama vs ChatGPT.

FAQ

Is Llama 4 Maverick free?

The weights are free to download under Meta's license, and you can chat with Maverick free on Llama.Online within the daily message allowance. Cloud APIs charge per token.

Can Maverick read images?

Yes. It accepts images together with text. In our chat you can attach one image per message while a Llama 4 model is selected. It cannot generate images.

Is Maverick the newest Llama?

It is one of the two newest, released with Scout on April 5, 2025. There is no Llama 5 as of October 2026; Meta's newer models are called Muse and are not part of our chat.

What does "17B-128E" mean in the model name?

It stands for 17 billion active parameters and 128 experts. Only part of the model works on each token, which keeps answers fast.

Does Maverick know about recent events?

No. Its training data ends in August 2024, and our chat has no web access, so verify anything time-sensitive yourself.