David Bridges
Meta has officially unveiled its first two innovative models from the highly anticipated multimodal Llama 4 suite: the Llama 4 Scout and Llama 4 Maverick. The Maverick model is touted as “the workhorse” among the two, showcasing exceptional capabilities in understanding both images and text, making it ideal for diverse applications such as general assistance and chat functionalities. In contrast, the smaller Scout model is designed to excel at tasks like multi-document summarization, efficiently parsing extensive user activities for personalized task recommendations, and reasoning over vast codebases. Additionally, Meta has teased the upcoming Llama 4 Behemoth model, which is claimed to be “among the world’s smartest large language models (LLMs).” CEO Mark Zuckerberg also hinted at a fourth model, Llama 4 Reasoning, which is expected to be revealed in the next month.
Both the Maverick and Scout models are now available for download from the official Llama website and Hugging Face, and they have been integrated into various Meta AI platforms, including popular messaging services like WhatsApp, Messenger, and Instagram DMs. This accessibility is expected to enhance user engagement and facilitate more efficient communication through advanced AI capabilities.
The Scout model impresses with its 17 billion active parameters and 16 specialized experts, providing an incredibly fast processing capability. According to Zuckerberg, it is not only natively multimodal but also boasts an industry-leading context length of nearly 10 million tokens, enabling it to run efficiently on a single GPU. On the other hand, the Maverick model also features 17 billion active parameters but is supported by 128 experts. Meta claims that it outperforms competitors such as GPT-4 and Gemini 2.0 across various benchmarks, including coding, reasoning, multilingual capabilities, long-context processing, and image recognition, proving its prowess in the AI landscape.
Zuckerberg has high expectations for the forthcoming Behemoth model, which is currently in training. He refers to it as “the highest performing base model in the world,” featuring a staggering 288 billion active parameters. Although it is still under development, anticipation is building as Meta gears up for its big AI developer conference, LlamaCon, scheduled for the coming weeks. This event is expected to unveil more exciting details about the Behemoth and the Reasoning model, further solidifying Meta’s position at the forefront of AI innovation.
Brandon Fletcher
Daniel Mercer
