Meta has unveiled Llama 4, the next generation of its open-weight language model family. The release includes several sizes, from a lightweight edge model to a large reasoning variant.
What is new
Llama 4 adds native multimodal capabilities, meaning it can process images, charts, and documents alongside text. Meta says reasoning and coding performance improved substantially over Llama 3, especially for longer contexts.
More flexible licensing
Meta has relaxed some commercial restrictions, making Llama 4 more attractive to startups and enterprises that want to build products without relying on third-party APIs. Very large platforms still face special licensing terms.
Availability
The models are available through Meta’s website, Hugging Face, AWS, Azure, and Google Cloud. Meta also published updated safety evaluations and a responsible use guide.