Meta Releases New Llama 4 AI Models With Multimodal Design
This article is for subscribers only.
Meta Platforms Inc. announced the release of its new Llama 4 artificial intelligence models, built on what the company says is one of the world’s most advanced large language models.
At the core of the new Llama 4 family is Llama 4 Behemoth, a two-trillion-parameter LLM that is still in training, with two distillations from it — dubbed Maverick and Scout — available right away for developers to build on and users to try via Meta’s apps or Meta.ai website. They come with the promise of being natively multimodal, meaning capable of working with a variety of media beyond text, and benchmark scores that Meta says puts them ahead of well-known competitors.