Llama 2

Meta Llama: open-weight, multimodal LLMs with 10M-token context, edge options, multilingual support, and state-of-the-art performance.

AI Assistant Updated 9 minutes ago

Visit Website

Llama 2's Top Features

Native multimodal models (text+vision) with early fusion and joint pre‑training on text, image, and video data

Industry‑leading long context: Llama 4 Scout supports up to 10 million tokens (trained at 256K) for strong length generalization

Mixture‑of‑Experts architecture: Scout (17B active/109B total, 16 experts) and Maverick (17B active/400B total, 128 experts)

Open‑weight and open‑source availability with 300M+ cumulative downloads and a permissive license for model output use

Extensive multilingual coverage: pre‑training across 200 languages with 10x more multilingual tokens than Llama 3

Edge/on‑device deployment: Llama 3.2 1B and 3B models run locally for private, low‑latency experiences

State‑of‑the‑art performance on STEM, coding, reasoning, multilingual, long‑context, and image benchmarks

Efficient training at scale: MetaP for stable hyper‑parameters, >30T token data mix, FP8 training on 32K GPUs achieving ~390 TFLOPs/GPU

Advanced tool use and reasoning, with small models outperforming peers on instruction following and tool-calling tasks

Broad range of model sizes from 1B to 405B parameters to fit mobile, server, and frontier workloads

Frequently asked questions about Llama 2

Meta Llama is an open-source family of advanced large language models with text and image understanding, reasoning, coding, and multilingual conversation, with Llama 3 and Llama 4 offering multimodal abilities, extended context, and state-of-the-art results.

They are openly available to developers, researchers, enterprises, and the community. Models can be downloaded from llama.com and Hugging Face and integrated across Meta’s products and partner platforms.

Recent releases include Llama 3.2 (text and vision models) and Llama 4 (Scout and Maverick), which add native multimodality, longer contexts (up to 10M tokens), and Mixture-of-Experts architectures for accuracy and efficiency.

Llama 3.2 lightweight models (1B and 3B) can run locally on devices, so data like messages or calendar items stays on-device unless shared intentionally, delivering fast, private responses without cloud dependence.

Across the Meta ecosystem: the Meta AI website, desktop and mobile apps, Ray‑Ban Meta glasses, and messaging platforms such as WhatsApp, Messenger, and Instagram Direct.

Maverick excels at multimodal tasks and benchmarks with 17B active and 400B total parameters, while Scout leads in context length up to 10M tokens and advanced retrieval/summarization over large document and code corpora.

Yes. Meta ships frequent major updates—with multiple Llama 4 releases planned for 2025—expanding capabilities in speech, reasoning, context length, and more.

Meta provides documentation, integration guides, and hosting options to help teams deploy Llama models, and actively encourages community feedback to improve capabilities.

Yes. Recent Llama releases support multilingual and multimodal (text and image) tasks, with ongoing work to broaden language coverage and add new modalities.

Meta promotes openness to drive innovation, collaborates with the community, and offers tools for safe, responsible deployment of LLMs.