Llama 4 Scout: Long-Context Llama Model, Free Chat

What Llama 4 Scout is and who makes it

Llama 4 Scout is a large language model from Meta, released on April 5, 2025 alongside Llama 4 Maverick. It is part of Llama 4, which is still the newest Llama generation as of October 2026, and its weights are downloadable under the Llama 4 Community License.

Scout reads text and images and answers in text. Like Maverick, it is a mixture-of-experts model: Meta's model card lists 109 billion parameters in total, 16 experts, and 17 billion parameters active for each token. Its standout feature is context length. Meta's card lists a context window of 10 million tokens, by far the longest in the Llama family, and Meta's launch post pointed to uses such as summarizing many documents at once and reasoning over large codebases. Meta also wrote that Scout fits on a single NVIDIA H100 GPU with Int4 quantization, which makes it much easier to host than Maverick.

Scout is one of the four models in our chat. Llama.Online is an independent site, not run by Meta; Scout runs on a cloud inference provider, and our site sends your messages to it.

Where it sits in the Llama family

Scout and Maverick share the same design, the same 12 officially supported languages and the same knowledge cut-off of August 2024. Maverick has 128 experts and 400 billion parameters in total, so it usually gives richer answers; Scout is lighter, cheaper to run and handles far longer inputs. Compared with text-only models such as Llama 3.3 70B, Scout adds image input and more languages. Meta's Llama Guard 4 safety classifier was itself built by pruning a Scout checkpoint. For every release in order, see Llama history.

What Scout is good at

Long documents. Reports, transcripts, contracts and long articles where you want a summary, a list of decisions or answers to specific questions.
Comparing several texts. Paste two versions of a document and ask what changed, or compare several product descriptions side by side.
Image questions. Meta describes Scout as strong at image grounding, which means linking parts of your question to specific regions of a picture. Ask it what a sign says, what is wrong in a screenshot, or what a chart shows.
Everyday chat and writing. Explanations, emails, outlines and rewrites, in any of the 12 supported languages: Arabic, English, French, German, Hindi, Indonesian, Italian, Portuguese, Spanish, Tagalog, Thai and Vietnamese.
Self-hosting. If you want a Llama 4 model on your own server, Scout is the realistic choice.

How to use Scout in our chat

Open the chat with the link below, or pick Llama 4 Scout from the model menu.
Paste the text you want to work with, or attach one image per message.
Ask a precise question: "List every deadline in this text with the person responsible" works better than "What is this about?"

Chat with Llama 4 Scout

The chat is free within a daily allowance of messages, currently 100 per day. Our summarizer tool is another quick way to shorten long text.

Limits to know about

Our chat trims long conversations. Our provider offers Scout 1,310,720 tokens of context, and the model card lists 10M, but our free chat sends only the most recent part of a long conversation and limits how much text one message can carry. For really large inputs, use the API or run Scout yourself.
Long context is not perfect recall. With very long inputs any model can miss or mix up details. Ask for quotes or section references so you can check.
Knowledge cut-off. Training data ends in August 2024, and there is no web access in our chat.
Mistakes happen. Scout can state wrong facts with confidence and can misread small or blurry text in images. Verify what matters.
Not the newest. Newer models, including Meta's own Muse family, are generally stronger at demanding coding and multi-step tasks. None of the Muse models are available in our chat.

Getting Scout officially

The weights are on Hugging Face as meta-llama/Llama-4-Scout-17B-16E-Instruct; accept the license on the model page to unlock the files. Ollama lists the Scout download at 67 GB, so you need a machine with a lot of memory. Through cloud providers it is inexpensive: OpenRouter lists it at $0.10 per million input tokens and $0.30 per million output tokens (OpenRouter list price, Oct 2026).

The Llama 4 Community License allows commercial use with conditions: show "Built with Llama" in your product, follow the acceptable use policy, and ask Meta for a separate license if your products had more than 700 million monthly active users. As a multimodal model, Scout falls under the policy's EU clause, which does not grant these rights to people or companies based in the EU, though end users of products built with it are not affected. Details are on our license page.

Privacy notes

Long documents often contain personal or confidential information. In our chat, what you paste is sent through our server and AI gateway to the inference provider to produce the answer; we do not store your message text, and your history stays in your browser. Still, remove names, account numbers and anything secret before pasting. For confidential material, a self-hosted Scout keeps the data on your own hardware.

How Scout compares

Choose Scout over Maverick when your input is long or your budget is tight, and Maverick when you want the richest answers; our Maverick vs Scout page goes into detail. Compared with Llama 3.3 70B, Scout is newer, reads images and covers more languages, as explained in Llama 3.3 vs Llama 4. Against newer open models from other companies, such as DeepSeek V4, Scout is older but lighter; see Llama vs DeepSeek.

FAQ

Can I really use 10 million tokens with Scout?

That is the figure on Meta's model card. Hosting providers usually offer less; ours lists 1,310,720 tokens, and our free chat trims long conversations further. Self-hosting at extreme lengths needs a great deal of memory.

Is Scout worse than Maverick?

It is smaller, so for nuanced writing or tricky questions Maverick is usually the better pick. For long inputs, quick answers and lower cost, Scout is often the smarter choice.

Does Scout read images?

Yes. Attach one image per message in our chat while Scout is selected. It describes and reasons about images but cannot create them.

Can I run Scout on my own computer?

Only on a machine with a lot of memory, since the download alone is about 67 GB according to Ollama. Smaller Llama models are easier to start with; see run Llama locally.