Meta introduces the Llama 4, now there are two new AI models, and there are two more
Meta released the first two models from its multimodal Camel 4 kit: Camel 4 Scout and Camel 4 Camel 4. The company said in a blog post that Maverick is the “workhorse” of both, good at comprehension of image and text for “general assistants and chat use cases”, while smaller model scouts can address “multi-file summary, extensive user activity for personalized tasks, and reasoning for a broad range of code.” The company also launched the Llama 4 Behemoth, the upcoming model “is the smartest LLM in the world” – CEO Mark Zuckerberg said we will hear about the fourth model, Llama 4 reasoning, “next month.”
Both Maverick and Scout are now available for download from the Llama website and Hugging Faces and have been added to Meta AI, including WhatsApp, Messenger and Instagram DMS.
Meta said the Scout has 17 billion active parameters and has 16 experts. “It's very fast, local multimodal, and has an industry-leading, almost unlimited 10 million token context length designed to run on a single GPU,” Zuckerberg said. “On the other hand, the Mavericks have 17 billion active parameters and 128 experts.” The company said it beats competitors like GPT-4O and Gemini 2.0 to coding, reasoning, multilingual, novel and image benchmarks, and compares with DeepSeek v3.1 in terms of reasoning and coding.
Zuckerberg has called the upcoming behemoth model “the world's highest performing fundamental model” with 288 billion active parameters training, the company said. It may not be here yet, but we will likely hear more inference models soon. Meta's large AI developer conference, Llamacon, has only been a few weeks now.
This article originally appeared on Engadget
