META DROPED LLAMA 4: What to know about the two new AI models

Meta presented its most new family of generative AI model On Saturday. You can give Llama 4 Testing Models now through the Meta AI website, and Llama 4 will soon power the many Meta Characteristics in the services of Instagram, WhatsApp and Messenger of the company.

Competition between Meta and other AI companies is becoming more and more intensive. Companies are working to build and release AI models capable of more complex tasks and sophisticated reasoning, without requiring huge quantities of computing power and money to execute. It is a complicated sweet place for shocks and Meta hopes that its most recent models will put it in front of competitors like Chatgpt and TwinsS

Now there are two models in the Llama 4 family: Scout and Maverick. They are models with open weights and multimodal ones, which means they can generate text, images and code. Open models like Meta mean that developers can get some idea of ​​how models are built. Llama 4 models are open weight models, which means that you can see how the model makes connections and how certain characteristics get more weight while learning. Openai announced earlier this month that it is developing Open weight For the first time.

The scout is the smallest family model designed to work on a GPU NVIDIA H100. SCOUT has a 10 million context window of the tokene and is a model of parameters of 17 billion, including 16 experts (subnet within the model, which allows it to perform the tasks more efficient). The scout has more than twice as big as the firing power of Llama 3who has 8 billion parameters. In general, the more parameters the model has, the more capable it is to achieve better results more quickly.

Maverick is an average model, Big Brother to Scout, including 17 billion parameters with 128 experts. Meta said, Deepseek V3S Deepseek still has the advantage when it comes to reasoning and encoding. CNET has not independently checked the Meta comparative analysis tests.

More information about the rest of the Llama 4 family, including a basic model called Behemoth and a Llama 4 reasoning model, is expected to come later this month, According to a video published by Executive Director Mark ZuckerbergS We will probably learn more about these models at Llamacon, the first annual AI developer conference of the company starting on April 29.

For more information, see what we know about a potentially self -app for meta AI and our review of The best AI chatbotS