What Meta builds
Meta makes Llama and publishes the weights, which is not a small thing. But the licence is called the Llama 4 Community License and it is not a standard open source licence: it is free up to 700 million monthly active users, and above that you have to ask Meta.
- Weights are published
- A custom licence, not open source
- Persian is not on the language list
Last checked: This is a reference page. It is re-checked against the vendor sources and updated when a new version ships. What changed
Model families
The Llama 4 line, two members with two different expert layouts.
Versions
| Versions | Released | Status | Model families |
|---|---|---|---|
| Llama 4 Maverick | current | Llama 4 | |
| Llama 4 Scout | current | Llama 4 |
Persian is not among the listed languages
This is the most important sentence on this page for a reader here. On the model card for both Llama 4 members, Meta lists exactly twelve languages: Arabic, English, French, German, Hindi, Indonesian, Italian, Portuguese, Spanish, Tagalog, Thai and Vietnamese. Persian is not one of them.
Notice how the list is written, because this differs from the Mistral case. Meta does not say including; it gives a closed list of twelve. So the absence of Persian here is not an inference. The model will probably produce Persian, but Meta has not claimed its quality and does not support it, and for work you have to hand a client those are not the same thing.
The licence is not open source, and that needs saying precisely
Llama is open source is repeated everywhere. The actual licence is the Llama 4 Community License and it carries three conditions a standard open source licence does not. First, if on the release date of that version you had more than 700 million monthly active users, you must request a separate licence from Meta. Second, you must prominently display Built with Llama. Third, any model you derive from it must have a name that begins with Llama.
For nearly every reader of this page the first condition is meaningless and the other two are real. Fine-tune the model and put your own name on it and you are in breach. The licence text also says your use must comply with trade compliance laws including export controls, and on that we neither advise nor describe any way around it.
Bigger model, smaller window
This is counterintuitive and it shows up in the table. Scout has 109 billion total parameters and a ten million token context window. Maverick has 400 billion total parameters and a one million token window, a tenth of Scout. Both activate 17 billion parameters per pass; what differs is the number of experts, 16 against 128.
So if your job is reading one very long document, the smaller model is the right choice. Anyone assuming the bigger number is always better gets this one wrong.
The knowledge cutoff, which is the oldest in this reference
Both Llama 4 models sit on August 2024. For comparison, Opus 5 sits on May 2026, about twenty one months ahead. For work tied to recent libraries or recent events this is Llama 4 biggest practical weakness, and it has nothing to do with the quality of the model.
A small thing that wastes your time
The Llama documentation moved from llama.com to developer.meta.com and the old address 301s. Old links work; deeper paths do not necessarily.
What it is good at
- The weights really are published, so the model runs on your own hardware and depends on no account
- Scout ten million token window, the largest figure any maker in this reference publishes
- Both models are multimodal and take image input as well as text
- The model card states total parameters, active parameters and expert count separately, which is transparency not everyone offers
Where it falls short
- Persian is not among the twelve languages Meta lists. Arabic is.
- The knowledge cutoff is August 2024, the oldest among the live models in this reference.
- The licence is not standard open source: it has a user ceiling, mandatory Built with Llama display, and derived model names must begin with Llama.
- Meta publishes no API price for these models, because the product is the weights and not a service. Your cost is hardware.
Our take
If you have to run the model yourself and you read very long documents, Scout with its ten million token window is a serious proposal. If your work is in Persian, this family is not the first choice and we do not recommend it.
Questions people actually ask
Does Llama know Persian
Meta lists twelve languages on the Llama 4 model card and Persian is not one of them, so Meta has not claimed Persian quality. The model may well produce Persian, but that is not the same as declared support.
Is Llama really open source
The weights are published but the licence is not standard open source. It is the Llama 4 Community License, with a 700 million monthly active user ceiling, mandatory Built with Llama display, and a requirement that derived model names begin with Llama.
Scout or Maverick
Scout has a ten million token window and Maverick has one million, even though Maverick is the larger model. For long documents, Scout. Both activate 17 billion parameters per pass.