Meta has launched Llama 4, a new series of AI models, on April 5, 2025, featuring Scout, Maverick, Behemoth, and L4emoth. These models focus on multimodal capabilities and efficiency via a mixture-of-experts (MoE) architecture, which allows them to perform tasks more effectively while cutting costs. Scout has 109 billion parameters and excels in processing lengthy documents, while Maverick, with 400 billion parameters, is designed for creative and multilingual applications. Behemoth is still in training and is expected to be Meta’s most powerful model, aimed at STEM tasks. Llama 4 is used in Meta’s apps like WhatsApp and Instagram, but its advanced features are currently limited to English users in the U.S. This launch underscores Meta’s dedication to open-source AI amid strong competition and regulatory challenges.
Meta has launched its latest AI series, Llama 4, on April 5, 2025. This new line includes four models: Scout, Maverick, Behemoth, and L4emoth. Designed for advanced tasks, these models focus on multimodal capabilities and are efficient in their computations. They use a mixture-of-experts (MoE) architecture, which allows them to manage tasks better by assigning special roles to specific components.
Currently, Scout and Maverick are available on Meta’s platforms and through partners like Hugging Face. Scout, with 109 billion parameters, is particularly good at handling long documents, managing a context window of up to 10 million tokens. On the other hand, Maverick features 400 billion parameters and is tailored for general assistant tasks including creative writing and multilingual support. Behemoth, still in training, is set to be the most powerful model from Meta, aimed at science, technology, engineering, and mathematics (STEM) applications.
Meta has also integrated Llama 4 into its AI assistant on WhatsApp, Messenger, and Instagram across 40 countries. However, its advanced features are currently only available to English-speaking users in the U.S. The company faces licensing restrictions in Europe due to regulatory concerns, needing special permits for large enterprises.
With the launch of Llama 4, Meta aims to stay competitive against powerful companies like OpenAI and DeepSeek from China. Internal tests show that while Llama 4 shows improvements in certain areas, it still trails behind some competitors like Google and Anthropic in reasoning capabilities. This new release marks a significant step in the evolution of the Llama ecosystem and highlights Meta’s commitment to advancing open-source AI amidst regulatory challenges.
In summary, Llama 4 represents an exciting advancement in AI technology and is a key part of Meta’s strategy to enhance its services while addressing the competitive landscape in the AI industry.
Source: Meta
Tags: Meta, Llama 4, AI Models, Multimodal AI, Open-source AI, Tech News
What is Llama 4’s 10M token context window?
Llama 4 has a large context window of 10 million tokens. This means it can remember and work with a lot of information from previous conversations, making it better at understanding complex topics and keeping track of details.
How does MoE architecture improve performance?
MoE stands for Mixture of Experts. This design lets Llama 4 use different specialized models for different tasks, helping it perform better and faster. It knows which expert to use based on the task, so it can provide more accurate answers.
Why is a larger context window important?
A larger context window helps Llama 4 hold onto more information at once. This is especially useful for long discussions or when the topic is complex, as it allows for more in-depth and coherent conversations.
Can Llama 4 learn from previous chats?
Yes, Llama 4 can build on previous conversations thanks to its large context window. This makes it able to provide responses that are relevant to earlier topics, creating a smoother and more engaging dialogue.
How does this affect everyday users?
With its advanced features, Llama 4 can enhance user experience in apps like chatbots and virtual assistants. Users can enjoy more meaningful interactions and receive better support for their inquiries or tasks.
