Meta released two llama 4 AI model

There is a commodity Announced Lama 4Now the newest AI Models Collection that strengthens the Meta AI assistant On the web And WhatsApp, Messenger and Instagram. Two new models that can be downloaded From Meta or Hugging faceLama 4 Scout- A small model that can be attached to a single NVIDIA H100 GPU ”and GPT-4O and Gemini 2.0 flash of Llama 4 Maverick. Meta says that she is still in the process of educating META CEO Llama 4 Behemoth Mark Zuckerberg says “The world’s highest performance basic model”.

According to Meta, Llama 4 SCOUT 10 million-based context window-a AI model of the AI ​​model-and Google’s Gemma 3 and Gemini 2.0 flash Lite models, as well as in the comparison of very broadly reported ”,“ a single Nvidia H100 GPU, in a single Nvidia H100 GPU ”. Meta, Openai’s GPT-4O and Google’s Gemini 2.0 flash makes similar allegations about the performance of the larger Maverick model, and its results are comparable to Deepseek-V3 in coding and reasoning tasks using “less than half of the active parameters”.

In the meantime, Lama 4 Behemoth has a total of 288 billion active parameters with a total of 2 trillion parameters. Although not yet published, Meta says that Behemoth can perform better than its competitors (in this case GPT-4.5 and Claude Sonnet 3.7) can perform better in “several STEM criteria”.

For Lama 4, Meta says that it has passed to the “mixture of experts” (MOE) architecture, which protects the resources using the parts of a model required for a specific task. The company plans to discuss its future plans for AI models and products at the Lamacon Conference on 29 April.

As in the past models, Meta calls Lama 4 collection as “open source ve, but was criticized for Lama license restrictions. For example, Lama 4 license Commercial institutions with more than 700 million active users per month require permission from Meta before using the models of open source initiative. Written in 2023 “Open source” category. “