Meta Yeni Llama 4 introduced artificial intelligence models

Meta announced the Llama 4 models, which brought the artificial intelligence family to a whole new level. Three new models carrying the names of Scout, Maverick and Behemoth officially introduced.

Meta states that it has trained these models with a large amount of text, visual and video data in order to give “wide visual understanding .. The company’s attack in this field accelerated after the Chinese artificial intelligence company Deepseek’s outdoor models have performed with a similar or better performance with Llama.

Scout, Maverick and Behemoth introduced

Scout and Maverick, one of the new models, are accessible via Meta’s official site and Huging Face. Behemoth, which has huge dimensions, is still in the education stage. Meta states that these models use a Moe – MoE – Mixture of Experts architecture for the first time. Thanks to this architecture, tasks are divided into specialized mini-models and processed more efficiently.

Scout stands out with its huge 10 million context window. This, to analyze long documents and images at the same time provides. Maverick has a parameter of 400 billion, but only 17 billion. Especially in areas such as creative spelling, multilingual understanding and image comparison, GPT-4O and Gemini can exceed 2.0, but it is still behind the new competitors, GPT-4.5, Claude 3.7 and Gemini 2.5 Pro.

Behemoth is about Two trillion total parameters and an artificial intelligence giant with 288 billion active parameters. In addition to standing in the tests performed in the field of STEM, even strong competitors such as GPT-4.5 leave behind from time to time.

Meta says that these new models are a turning point in the LLAM ecosystem. According to the company’s statement, these models are only one start.

OTHER CONTENTS RELATED TO META: