Meta Launches Llama 4 AI Models
Meta has unveiled two new AI models, Llama 4 Scout and Llama 4 Maverick, calling them its most advanced systems yet. Both are open source and part of a broader push to make AI tools more accessible. These models are multimodal, meaning they can understand and work across text, images, audio, and video. Users can try them now through Meta’s apps or the Meta.ai website.
At the center of the new Llama 4 lineup is the upcoming Llama 4 Behemoth, a massive two-trillion-parameter model still in training. Meta says it will serve as a “teacher” for future versions and could power AI agents capable of more complex tasks like web surfing and advanced reasoning. The company is betting big on these tools, adopting a Mixture of Experts (MoE) architecture to improve efficiency and cut costs—an approach already used by some of its Chinese competitors like DeepSeek.
Mark Zuckerberg said Meta’s goal is to build the world’s leading AI and make it open source “so that everyone in the world benefits.” The company plans to spend up to $65 billion this year to expand its AI infrastructure and will host its first LlamaCon conference in late April. A standalone Meta AI app is also expected soon.
