Meta Releases Llama 4 Scout & Maverick Models, Teases ‘Behemoth’ AI Powerhouse
Meta has officially launched two new models from its Llama 4 AI suite — Llama 4 Scout and Llama 4 Maverick — both of which are now available via the Llama website and Hugging Face. The announcement was made in a blog post published on Meta’s official AI website.
- Maverick: Built for general assistant and chat use cases, with strong capabilities in both image and text understanding.
- Scout: A lightweight model aimed at multi-document summarization, deep reasoning over codebases, and personalization tasks.
According to Meta CEO Mark Zuckerberg, Scout has 17B active parameters with 16 experts and features a massive 10 million token context window, while Maverick uses 128 experts and outperforms competitors like GPT-4o and Gemini 2.0 in reasoning, coding, and multilingual benchmarks.
Meta also teased Llama 4 Behemoth, which is still in training. The company says it contains 288B active parameters and will be “the highest performing base model in the world.” A fourth model, Llama 4 Reasoning, will be announced in the coming weeks — possibly during LlamaCon, Meta’s AI developer conference.
Sources:
- Meta AI Blog – Llama 4 Announcement
- Hugging Face – Meta AI Models
- Mark Zuckerberg (via Meta press release)