Meta has Google and OpenAI squarely in its sights with new Llama ...
Meta is introducing two new models in the Llama 4 series — Scout and Maverick. Early tests have shown that both models are surpassing the competition. Scout is designed for handling large documents, complex requests, and extensive codebases, while Maverick excels in processing both text and visuals, making it perfect for smart assistants and chat interfaces. These models are currently live on Llama.com and available through partners such as Hugging Face.
Enhanced AI Integration
Meta AI is expanding its reach by becoming accessible on WhatsApp, Messenger, and Instagram in 40 countries (limited to the U.S. and English-only for now). The company is starting its multimodal Llama 4 series with two innovative models: Scout and Maverick, each offering unique capabilities that outperform competitors.
In a recent announcement, Meta shared that the new Scout and Maverick models have demonstrated superior performance compared to other models in various tests. Maverick, known for its versatility in handling text and visuals, is particularly well-suited for applications like smart assistants and chat interfaces. On the other hand, Scout is more lightweight and agile, making it ideal for tasks such as navigating complex documents and intricate codebases.
Technical Advancements
Llama 4 Scout and Maverick are now available for purchase on Llama.com and through Meta's partners, including Hugging Face. Meta has also integrated these new models into its Meta AI assistant, which is gradually being rolled out on platforms like WhatsApp, Messenger, and Instagram in multiple countries. However, the advanced multimodal features are currently limited to the U.S. and English-speaking users.
Meta's deployment of the Llama 4 series signifies a significant advancement under the hood, with the incorporation of a Mixture of Experts (MoE) setup for the first time. This setup enhances the efficiency and speed of tasks, both during model training and in responding to queries.
The MoE approach involves breaking down complex tasks into smaller segments and assigning them to specialized mini-networks. Scout is equipped with 17 billion active parameters distributed across 16 expert modules, outperforming several industry benchmarks while maintaining a lean profile that runs smoothly on a single Nvidia H100 GPU. Its standout feature includes a massive context window capable of handling up to 10 million tokens, enabling effective processing of both textual and visual data on a large scale.
Maverick, with 17 billion parameters spread across 128 expert networks, has demonstrated remarkable performance comparable to industry-leading models but with significantly fewer active parameters.
Future Developments
Meta has teased the upcoming release of the Llama 4 Behemoth, currently in the training phase, expected to be one of the most advanced LLMs available. Behemoth boasts 288 billion active parameters across 16 experts, with a total parameter count nearing two trillion.
Despite its capabilities, Maverick falls slightly short compared to top-tier models like Google's Gemini 2.5 Pro and OpenAI's GPT-4.5 in terms of overall performance.
Overall, Meta's ongoing enhancements in AI integration across its platforms promise more sophisticated logic systems that deliver sharper responses, enhanced image generation, and more precise ad targeting.
