Meta Unveils Llama 4: Two AI Models Out Now, Two More in the Pipeline
Menlo Park, CA – April 5, 2025 – Meta has officially launched Llama 4, the latest chapter in its open-source AI saga, introducing two immediately available models—Llama 4 Scout and Llama 4 Maverick—while teasing two more, Llama 4 Behemoth and Llama 4 Reasoning, slated for release soon. Announced on Saturday, this move underscores Meta’s aggressive push to lead the AI race, blending cutting-edge tech with its signature open-access ethos.
Llama 4 Scout and Maverick hit the ground running, available for download on Llama.com and Hugging Face, and already powering Meta AI across WhatsApp, Messenger, Instagram, and the web. Scout, a lean 17-billion-parameter model with 16 experts, is built for efficiency, fitting on a single Nvidia H100 GPU. It boasts a groundbreaking 10-million-token context window—think digesting entire books or vast codebases in one go—and excels at tasks like multi-document summarization and personalized activity parsing. Maverick, also 17 billion active parameters but with 128 experts and 400 billion total, is the heavier hitter, tackling image and text understanding for chat and assistant use cases. Meta claims it outpaces GPT-4o and Gemini 2.0 Flash on coding, reasoning, and multilingual benchmarks, rivaling DeepSeek V3.1 in spots.
Both models lean on a mixture-of-experts (MoE) architecture, a first for the Llama family, making them computationally nimble while packing multimodal punch—text and images in, text out. They’re distilled from the still-training Llama 4 Behemoth, a beast with 288 billion active parameters and nearly 2 trillion total, which Meta CEO Mark Zuckerberg calls “the highest-performing base model in the world.” Behemoth’s internal benchmarks reportedly top GPT-4.5 and Claude 3.7 Sonnet in STEM tasks, though it’s not out yet. The fourth model, Llama 4 Reasoning, promises advanced logic chops and is expected to debut by May, with more details likely at Meta’s LlamaCon on April 29.
Meta’s doubling down on openness, but with caveats: the EU’s locked out due to regulatory gripes, and firms with over 700 million monthly users need special permission. Still, Zuckerberg’s betting big—$65 billion this year on AI infrastructure—to make Llama the go-to for developers and businesses. Early buzz on X calls it a game-changer, with Scout’s context length and Maverick’s versatility stealing the show. As the AI landscape heats up, Llama 4’s rollout could shift the balance—watch this space.