Meta unveiled the LLAMA 4 series artificial intelligence models • DigiKala Meg

Last night, Meta unveiled its new series of artificial intelligence models, Llama 4. These models are to be used as a smart meta assistant in applications such as WhatsApp, Messenger and Instagram. To date, the Meta has made two models of this collection available and the other two models, which are more powerful, have not yet been unveiled.

The LLAMA 4 Scout, with 109 billion parameters, is a smaller option that can even run with the Nvidia H100 GPU. In contrast, the LLAMA 4 Maverick model with 400 billion parameters, with a level with Gpt-4O and Gemini 2.0 Flash. These two models are now ready to download from the Meta website or the Hugging Face website.

According to Meta, the Llama 4 Behemoth model is still in the training phase. CEO Mark Zuckerberg has described the model as “the world’s most powerful base model”. Also, the fourth model, called the Llama 4 Reasoning, is set to be released next month. The Scout model utilizes a 10 million tunes background and has been able to overcome models like the Gemma 3, Gemini 2.0 Flash-Lite and Mistral 3.1, all of which are open. Meta also claims that Maverick performs superior to the Gpt-4O and Gemini 2.0 Flash and competes with Deepseek-V3 in the field of coding and reasoning, using less than half of its active parameters.

The Behemoth model, with 288 billion active parameters and a total of 2 trillion parameters, has not yet been released. However, Meta predicts that the model in the STEM benchmarks will surpass competitors like the Gpt-4.5 and Claude Sonnet 3.7. Meta has come to the “MOE) architecture for the development of LLAMA 4. In this way, different parts of the model are specialized in the tasks. These four models are also multidisciplinary and capable of processing and producing text, image and video, though no mention of sound processing.

Meta still calls the collection “open -minded”, but the permission restrictions have been criticized. For example, business companies with over 700 million active users of the month must obtain permission from meta before using these models. Last year, the OSI announced that it would remove meta models from the open source.

Meta is scheduled to provide more details about the future plans of these models and artificial intelligence products at the Llamacon Conference, which will take place on April 29. This event will be an opportunity for those interested in technology to get acquainted with the company’s future orientations. Given the increasing competition in the field of artificial intelligence, the LLAMA 4 supply can be an important step in this field. The company hopes to reinforce these models in the new technology market and improve its experiences on various platforms.

Source: Engadget