{"id":690,"date":"2025-04-07T18:07:52","date_gmt":"2025-04-07T18:07:52","guid":{"rendered":"https:\/\/www.techrepublic.com\/?p=4301356"},"modified":"2025-04-07T18:07:52","modified_gmt":"2025-04-07T18:07:52","slug":"meta-unveils-llama-4-ai-series-featuring-new-expert-based-architecture","status":"publish","type":"post","link":"https:\/\/blog.tecnoartesanos.com\/index.php\/2025\/04\/07\/meta-unveils-llama-4-ai-series-featuring-new-expert-based-architecture\/","title":{"rendered":"Meta Unveils Llama 4 AI Series Featuring New Expert-Based Architecture"},"content":{"rendered":"
\"Llama
Image: Meta<\/figcaption><\/figure>\n

Meta unveiled on April 5 its new AI model series: Llama 4, which includes Llama 4 Maverick and Llama 4 Scout, tailored for conversation and processing large files, respectively, along with an unreleased \u201cteacher\u201d model called Llama 4 Behemoth.<\/p>\n

Llama 4<\/a> is Meta\u2019s first series to adopt a \u201cmixture of experts (MoE) architecture.\u201d This approach activates only select parts of the neural network, referred to as the \u201cexperts,\u201d to handle specific subtasks. The task will be broken down into subtasks and each routed to the most appropriate experts, improving resource efficiency.<\/p>\n

<\/span>What are the specifics about Llama 4 Maverick and Scout?<\/span><\/h2>\n

Llama 4 Maverick features 128 experts and 17 billion active parameters, which represent the portion of a model\u2019s knowledge used to process a given input. Meta describes it as the \u201cproduct workhorse model for general assistant and chat use cases,\u201d specialising in image interpretation and creative writing.<\/p>\n

Interestingly, Mark Zuckerberg\u2019s company boasts that Maverick offers \u201ca best-in-class performance to cost ratio\u201d when it comes to conversations. Cost has been playing on the minds of AI giants since the surprise release of DeepSeek in January<\/a>, which took only $5.6 million to train<\/a>.<\/p>\n

SEE: Meta\u2019s $800M Offer To Chip Startup Was Rejected \u2014 Here\u2019s Why<\/a><\/strong><\/p>\n

However, AI experts have noticed that the version of Llama 4 Maverick published on LMArena, which ranks major large language models across various tasks, is \u201coptimized for conversationality<\/a>\u201d and performs differently<\/a> from the publicly available version. This suggests that Meta submitted an altered version to LMArena that would rank higher on its leaderboard.<\/p>\n

Llama 4 Scout also has 17 billion active parameters and just 16 experts, but Meta says it is the \u201cbest multimodal model in the world in its class.\u201d It has an unusually large context window of 10 million tokens, which represent the amount of information it can process in a prompt, so it performs well when summarising large documents and in sequential reasoning.<\/p>\n

Meta says that both Scout and Maverick are its \u201cbest yet\u201d due to being distilled from Llama 4 Behemoth, with a whopping 28 billion active parameters and 16 experts. While it already ranks highly on LMArena, it is still being trained and has not been released. According to The Information<\/a>, the Llama 4 announcement was delayed at least twice due to the models underperforming in technical benchmarks and conversationality.<\/p>\n

<\/span>How can you access LLama 4 Maverick and Scout?<\/span><\/h2>\n

Scout and Maverick can be downloaded on Llama.com and Hugging Face, or used through the Meta AI chatbots in WhatsApp, Messenger, and Instagram in 40 countries. Multimodal features can only be used in the U.S. and in English, currently.<\/p>\n

Some partners have already announced integrations; developers can build and deploy AI applications with the Llama 4 models in Microsoft\u2019s Azure AI Foundry and Azure Databricks.<\/p>\n