Meta has announced the release of Llama 4its new collection of AI models that now feed Meta Ai on the web And in WhatsApp, Messenger and Instagram directly. Both models, also available for download Meta Or Face Now are Llama 4 Scout, a small model capable of “adapting in a single NVIDIA H100 GPU” and LLAMA 4 Maverick, which is more like GPT-4O and Gemini 2.0 Flash. And the company says it is training the giant of the Llama 4, which Meta CEO Mark Zuckerberg says on Instagram is “already the most efficient basic model in the world”.
According to Meta, Scout has a context window of $ 10 million – the working memory of an AI model – and beats Gemma 3 and Gemini 2.0 models from Gemini 2.0 from Google, as well as the Mistral 3.1 Open Source, “on a wide range of widely reported references”, while “going up in a single Nvidia H100 GPU”. He makes similar affirmations on the performance of his larger Maverick model compared to the GPT-4O of Openai and Gemini 2.0 of Google, and says that its results are comparable to Deepseek-V3 in coding and reasoning tasks using “less than half of the active parameters”, or the variables that guide the behavior of AI models.
Meanwhile, the Llama 4 giant has 288 billion active parameters with 2 billions of parameters in total. The company again claims that Behemoth can surpass its competitors, in this case GPT-4.5 and Claude Sonnet 3.7, “on several STEM benchmarks”.
For Llama 4, Meta says that it went to a “mixture of experts” (MOE), an approach that keeps resources using only the parts of a necessary model for a given task. The company plans to discuss future plans for AI models and products in Llamacon, which takes place on April 29.
As with her previous models, Meta calls the Llama 4 collection “open-source”, although she was criticized for the less than open requirements of her licenses. For example, the LLAMA 4 license requires commercial entities with more than 700 million monthly active users to request a META license before using its models, that the open source initiative Written in 2023 The withdrawal of “Open Source” category. “
