Meta has unveiled its latest advancements in its Llama series. The upgrades for Llama 4 bring three new AI models called Scout, Maverick, and Behemoth.

These new iterations could give Meta a boost in the AI technology industry. The models were developed with a focus on processing vast amounts of data, and trained extensively on unlabeled text, images, and audio.

Both Scout and Maverick are now available for use, via Llama’s official website and the Hugging Face platform. Scout is a multimodal model, capable of processing diverse data types such as text, image, and audios simultaneously. This makes it particularly effective in tasks like document summarization and intricate reasoning across large databases. Meta asserts that Scout can analyze texts of over five million words instantly, making it invaluable for sifting through detailed research papers, meetings, etc.

In contrast, Maverick operates as a general-purpose assistant that excels in understanding both text and images. While it boasts a greater volume of parameters and a more extensive knowledge base than Scout, it falls short in contextual memory. This limitation means that while it handles a broader range of tasks, it may struggle in recalling lengthy conversations.

Behemoth, the company's largest and most advanced model, does not have a release date. Meta says that it is still training the AI. But, it says that Behemoth will merge the strengths of both Scout and Maverick, offering robust knowledge combined with extended contextual understanding. Internally, Meta claims that Behemoth has outperformed several leading AI competitors including OpenAI's GPT-4.5, Claude 3.7 Sonnet and Google Gemini 2.0 Pro in its internal benchmarks on STEM-related tasks, including complex problem-solving.

Meta is set to integrate the new Llama 4 models into its apps: WhatsApp, Facebook Messenger, and Instagram. The social media giant will launch the new AIs in 40 countries, but users in Europe won't be able to access them due to EU regulations.

Source: Tom's Guide