Meta launches Llama 4 AI model suite, powering 'Meta AI' across platforms
Meta has recently introduced its latest AI model suite, Llama 4, designed to enhance the functionalities of the "Meta AI" assistant on popular platforms like WhatsApp, Messenger, and Instagram. This suite comprises two key models, namely Llama 4 Scout and Llama 4 Maverick, both of which can be accessed for download on Meta's official website and the Hugging Face platform.
Introducing Llama 4 Behemoth
Additionally, Meta is actively developing the Llama 4 Behemoth model, which has been hailed by CEO Mark Zuckerberg as "the world's most powerful foundational model." Despite not being available yet, this model is reported to boast an impressive 288 billion effective parameters out of a total of one trillion. Meta asserts that it has the potential to surpass existing models like GPT-4.5 and Claude Sonnet 3.7 in various STEM assessments.
Advanced Architecture
The Llama 4 suite is built on a sophisticated "Mixture of Experts" architecture, which optimizes efficiency by activating only the relevant components of the model for specific tasks. While Meta positions Llama 4 as an open-source solution, the licensing terms have faced scrutiny. Companies with a user base exceeding 700 million are required to obtain Meta's authorization to leverage these models, a stipulation that has prompted the Open Source Initiative to exclude Meta's models from the "open source" category in a recent 2023 report.
Upcoming Announcements
Meta is gearing up to reveal more insights into its AI strategies at the upcoming LlamaCon conference scheduled for April 29. This event is expected to shed light on the company's future endeavors in AI development.
