Llama 4 Scout / Maverick

weights

open(downloadable on HF)

data

closed

code

partial(inference/utils via llama-models, no training code)

license

Llama-4-Community(non-OSI - 700M MAU commercial cap, acceptable-use policy with an EU-domicile exclusion for the multimodal models, Llama naming requirement for derivatives)

Llama 4's weights download from Hugging Face once a user accepts Meta's Llama 4 Community License and submits a contact form. The license caps commercial use at 700 million monthly active users and requires derivatives to carry the Llama name, and its acceptable-use policy withholds the multimodal models from individuals and companies based in the EU. That cap still permits commercial use, unlike Mistral's Non-Production terms or CC-BY-NC, which forbid it outright.

• https://ai.meta.com/blog/llama-4-multimodal-intelligence/ recorded 2026-08-13

  Now answers HTTP 400 to a plain request. It was originally cited as a corroborating launch post for the release.

• https://api.github.com/repos/meta-llama/llama-models recorded 2026-08-13

  `private: false`, `archived: false`, description "Utilities intended for use with Llama models", license `NOASSERTION`. Utilities and inference helpers, no pretraining pipeline, which is what holds `code` at partial.

• https://huggingface.co/api/models/meta-llama/Llama-4-Scout-17B-16E recorded 2026-09-29

  Hub API record for the Scout base checkpoint: gated is manual and private is false, the file list runs "model-00001-of-00050.safetensors" through fifty shards, the license tag is "license:other", and the gating prompt opens with the "LLAMA 4 COMMUNITY LICENSE AGREEMENT" text effective "April 5, 2025".

• https://huggingface.co/blog/llama4-release recorded 2026-09-29

  Hugging Face release post: "Both Llama 4 Maverick and Llama 4 Scout model weights are available directly on the Hugging Face Hub under the" meta-llama organization, and "This includes both base and instruction tuned variants", behind the condition "You need to accept the license terms on the model card before accessing the weights".

• https://huggingface.co/meta-llama/Llama-4-Scout-17B-16E recorded 2026-09-29

  Model card carries the Llama 4 Community License in full, including the cap "greater than 700 million monthly active users in the preceding calendar month, you must request a license from Meta" and the naming rule "include “Llama” at the beginning of any such AI model name", and describes the corpus without releasing it: "pretrained on ~40 trillion tokens and Llama 4 Maverick was pretrained on ~22 trillion tokens of multimodal data from a mix of publicly available, licensed data and information from Meta’s products and services".

• https://pypi.org/pypi/llama-models/json recorded 2026-09-29

  PyPI record for llama-models, the package built from Meta's meta-llama/llama-models repository: summary "Llama models", and a README whose code is download and inference tooling ("Example scripts are available in", with the Llama 4 scripts running "models.llama4.scripts.chat_completion"). Searched the README for a pretraining recipe or training script: none, only inference and quantized-inference instructions.

• https://royfactory.net/posts/ai/202512/meta-llama4-open-weights-scout-maverick-license/ recorded 2026-09-29

  Third-party summary of the release and its license, noting the "major threshold clause (700M MAU) you must review before production use"; corroborates the open-weights reading and establishes no dimension on its own.

• https://www.llama.com/llama4/use-policy/ recorded 2026-09-29

  Llama 4 Acceptable Use Policy (served from dev.meta.ai after redirect): "With respect to any multimodal models included in Llama 4, the rights granted under Section 1(a) of the Llama 4 Community License Agreement are not being granted to you if you are an individual domiciled in, or a company with a principal place of business in, the European Union" and "This restriction does not apply to end users of a product or service that incorporates any such multimodal models".

Adoption is measured as Hugging Face downloads summed across the five declared artifacts spanning Llama 3.1 and Llama 4. The older Llama-3.1-8B checkpoint still accounts for most of that reach, even though the capability reading is taken on Llama 4 Maverick -- reach and capability sit on different releases of the same family.

• https://huggingface.co/api/models/meta-llama/Llama-3.1-405B recorded 2026-10-07

  89,695 downloads in the trailing 30 days for meta-llama/Llama-3.1-405B

• https://huggingface.co/api/models/meta-llama/Llama-3.1-70B recorded 2026-10-07

  41,198 downloads in the trailing 30 days for meta-llama/Llama-3.1-70B

• https://huggingface.co/api/models/meta-llama/Llama-3.1-8B recorded 2026-10-07

  637,733 downloads in the trailing 30 days for meta-llama/Llama-3.1-8B

• https://huggingface.co/api/models/meta-llama/Llama-4-Maverick-17B-128E recorded 2026-10-07

  189 downloads in the trailing 30 days for meta-llama/Llama-4-Maverick-17B-128E

• https://huggingface.co/api/models/meta-llama/Llama-4-Scout-17B-16E recorded 2026-10-07

  3,641 downloads in the trailing 30 days for meta-llama/Llama-4-Scout-17B-16E

Llama 4 Maverick, the strongest released member of the family, scores 10 on Artificial Analysis's Intelligence Index, below the median of 12 for open-weight non-reasoning models of similar size. Llama 4 Scout scores lower, and the larger Behemoth model was never released.

• https://artificialanalysis.ai/models/llama-4-maverick recorded 2026-09-29

  Artificial Analysis page for Maverick on Intelligence Index v4.3.2: "Llama 4 Maverick scores 10 (estimated) on the Artificial Analysis Intelligence Index, placing it below average among other open weight non-reasoning models of similar size (median: 12)", ranked 25th of 45 in the model summary.

• https://artificialanalysis.ai/models/llama-4-scout recorded 2026-09-29

  Artificial Analysis page for Scout, the family's other released model: "Llama 4 Scout scores 8 (estimated) on the Artificial Analysis Intelligence Index", below Maverick's 10, so Maverick remains the family's strongest reading.