{"id":22828,"date":"2025-04-08T15:14:21","date_gmt":"2025-04-08T12:14:21","guid":{"rendered":"https:\/\/forklog.com\/en\/meta-unveils-new-series-of-ai-models-llama-4\/"},"modified":"2025-04-08T15:14:21","modified_gmt":"2025-04-08T12:14:21","slug":"meta-unveils-new-series-of-ai-models-llama-4","status":"publish","type":"post","link":"https:\/\/forklog.com\/en\/meta-unveils-new-series-of-ai-models-llama-4\/","title":{"rendered":"Meta Unveils New Series of AI Models: Llama 4"},"content":{"rendered":"

Meta Corporation has released<\/a> a new lineup of open AI models, Llama 4. According to internal tests, these models outperform competitors across various benchmarks. <\/p>\n

The series is anchored by Llama 4 Behemoth, a large language model (LLM) with 2 trillion parameters. It is currently in the training phase and has not yet been released. Two of its multimodal distillations<\/span>\u2014Maverick and Scout\u2014are available to developers and users.<\/p>\n

Meta AI assistant, available in various company products like WhatsApp, Messenger, and Instagram, has already been updated to use Llama 4 in 40 countries. However, multimodal features are currently available only in the United States. <\/p>\n

It is claimed that Behemoth, the LLM teacher of the other two models, surpasses GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro in STEM-oriented benchmarks<\/span> like MATH-500 and GPQA Diamond. <\/p>\n

\n

\u201cThis is just the beginning for the Llama 4 collection. We believe the most intelligent systems should be capable of performing generalized actions, naturally communicating with people, and solving complex tasks they have not encountered before. Empowering Llama with super capabilities in these areas will lead to the creation of better products for people on our platforms and expand developers’ ability to innovate in the next major consumer and business sectors,\u201d the company announcement states. <\/p>\n<\/blockquote>\n

New Architecture<\/strong><\/h2>\n

Llama 4 is the first series of models to use the Mixture of Experts (MoE)<\/span> architecture. Maverick has 128 \u201cexperts\u201d and 400 billion total parameters, but only 17 billion are active. Scout’s figures are 16, 109 billion, and 17 billion, respectively. <\/p>\n

\"Meta
Characteristics of neural networks from the Llama 4 lineup. Data: Meta.<\/figcaption><\/figure>\n

According to the company’s internal tests, Maverick outperforms models like GPT-4o and Gemini 2.0 in some programming, reasoning, language support, long contexts, and image tests. However, the neural network falls short of the more powerful and modern Gemini 2.5 Pro from Google, Claude 3.7 Sonnet from Anthropic, and GPT-4.5 from OpenAI.<\/p>\n

Maverick is better suited for use as a general assistant and chat. Scout’s strengths lie in document summarization and reasoning over large databases. The latter can operate on a single Nvidia H100 graphics processor, while Maverick requires a Nvidia H100 DGX system or its equivalent.<\/p>\n

Controversy Surrounding Llama 4<\/strong><\/h2>\n

Maverick secured second place in the LLM Arena\u2014a test where people compare the performance of various models and form a \u201cuser\u201d ranking. <\/p>\n

\"Meta
AI model rankings according to LLM Arena data. Data: LLM Arena.<\/a><\/figcaption><\/figure>\n

Several researchers noted<\/a> that a specially optimized version of Maverick, unavailable to developers, participated in the tests. The version for LLM Arena uses more emojis and provides unusually long responses. <\/p>\n

\n

Okay Llama 4 is def a littled cooked lol, what is this yap city pic.twitter.com\/y3GvhbVz65<\/a><\/p>\n

\u2014 Nathan Lambert (@natolambert) April 6, 2025<\/a><\/p><\/blockquote>\n