Meta has introduced two new AI models under its Llama 4 lineup, now powering the Meta AI assistant across platforms like WhatsApp, Messenger, Instagram, and the web. The company claims that these models outperform rivals from OpenAI and Google on various industry benchmarks. The two released models are Llama 4 Scout and Llama 4 Maverick, each offering distinct advantages in terms of size, efficiency, and performance.
Llama 4 Scout is a smaller model designed to run efficiently on a single Nvidia H100 GPU. Despite its compact design, Meta says it delivers better results than Google’s Gemma 3, Gemini 2.0 Flash-Lite, and the open-source Mistral 3.1, across multiple benchmarks. With a 10-million-token context window, Scout offers strong performance without requiring extensive computing power.
Llama 4 Maverick, the more advanced of the two, is positioned as a competitor to OpenAI’s GPT-4o and Google’s Gemini 2.0 Flash. Meta states that Maverick matches or surpasses these models in coding and reasoning tasks while using fewer active parameters. This makes Maverick a more efficient yet powerful option for complex applications.
Meta is also training a third model, Llama 4 Behemoth, which has not yet been released. It boasts 288 billion active parameters and a total of 2 trillion parameters. According to Meta CEO Mark Zuckerberg, Behemoth is expected to be the top-performing base model globally and is already surpassing models like GPT-4.5 and Claude Sonnet 3.7 in several STEM-focused benchmarks.
All three models use a “mixture of experts” (MoE) architecture, which increases efficiency by activating only parts of the model needed for specific tasks. Meta plans to share more details about its AI strategy and upcoming innovations at its LlamaCon conference on April 29.
Though labeled as open-source, Llama 4’s license has faced criticism. The license restricts commercial use for companies with over 700 million monthly users unless they get prior permission from Meta, a clause some argue disqualifies it from being fully open-source.
