Meta Deploys New Llama 4 Models and AI Agents

Meta has unveiled its next-generation Llama 4 series of large language models. Concurrently, its subsidiary Manus AI has launched new AI agents on Telegram, signaling a strategic push toward multi-agent systems and ecosystem tools ahead of an expected integration with WhatsApp.

- The Llama 4 family includes three models of varying sizes: the 109 billion-parameter "Scout," the 400 billion-parameter "Maverick," and the yet-to-be-released 2 trillion-parameter "Behemoth." Both Scout and Maverick utilize a Mixture-of-Experts (MoE) architecture, a first for the Llama series, which enhances efficiency by only activating a subset of parameters for any given task. - Llama 4 marks Meta's first natively multimodal model series, capable of processing text, images, and video as combined inputs from the start. This "early fusion" technique allows the model to be pre-trained on a mix of data types, enabling a more unified understanding of interleaved text and images. - The Llama 4 Scout model features a context window of up to 10 million tokens, a significant increase designed for tasks requiring analysis of very long documents or codebases. The Maverick model, which is used in Meta's own applications like Facebook and Instagram, is positioned to compete with models like GPT-4o and Gemini 1.5. - The acquisition of Manus AI is a key part of Meta's broader strategy to develop universal AI assistants that can perform complex, multi-step tasks. This aligns with the concept of multi-agent systems, where different AI agents collaborate to achieve a common goal, a field advanced by frameworks like MetaGPT. - Manus AI agents on Telegram are designed to go beyond simple chatbot interactions by maintaining long-term memory of user preferences and connecting to external tools like Gmail and Calendar. Users can interact with these agents via text, voice, or images to perform tasks like creating documents, processing data, or conducting research. - The decision to launch Manus AI agents on Telegram before Meta's own WhatsApp may be a strategic move to test the technology on a platform not directly tied to the Meta brand. This allows for public testing while the acquisition of Manus by Meta is still under review by some international regulators. - Meta is not only deploying these agents in messaging apps but is also integrating them into hardware like the Ray-Ban Meta smart glasses. This points to a future where AI agents can interact with the user's environment to provide real-time assistance.

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.