Particle.news
Download on the App Store
8 ARTICLES
·
3w ago

Meta Unveils Llama 4 AI Models with Multimodal and Efficiency Breakthroughs

The release includes Scout and Maverick models, while the larger Behemoth remains in training, showcasing Meta's push for open-source AI innovation.

Image
Image
Image
Image

Overview

  • Meta has launched the Llama 4 family, introducing Scout and Maverick models, with the larger Behemoth still in development.
  • Llama 4 models utilize a mixture-of-experts (MoE) architecture, improving computational efficiency by activating only specialized parameters for tasks.
  • Scout features a groundbreaking 10 million token context window for long-context tasks, while Maverick excels in reasoning and coding benchmarks.
  • The models are multimodal, capable of processing text, images, and video, and are available for download under Meta's open-source license, excluding EU users due to regional data laws.
  • Meta plans to showcase its advancements, including the Llama 4 family, at its inaugural LlamaCon AI conference on April 29, 2025.