Sign up to save tools and stay up to date with the latest in AI
bg
bg
1

Mark Zuckerberg says Meta's latest Llama models put open-source AI in the driving seat - SiliconANGLE

Apr 07, 2025 - siliconangle.com
Meta Platforms Inc. has unveiled its latest Llama 4 series of large language AI models, claiming they are among the most powerful open-source models available. The series includes Llama 4 Scout, a small model that fits inside a single Nvidia H100 GPU, and Llama 4 Maverick, a larger model comparable to OpenAI’s GPT-4o and Google’s Gemini 2.0 Flash. Both models have 17 billion active parameters and are designed to handle various tasks, with Maverick excelling in creative writing and problem-solving. Meta is also developing Llama 4 Behemoth, which will have 288 billion active parameters and aims to be the highest-performing base model globally.

Meta emphasizes the efficiency and safety of the Llama 4 models, using a “Mixture-of-Experts” system to optimize performance and energy use. The models are designed to be more balanced and less biased, providing nuanced responses to sensitive questions. Meta CEO Mark Zuckerberg highlighted the company’s commitment to open-source AI, aiming to make it universally accessible. The Llama 4 models are available through Meta’s platforms and are expected to be a focal point at the upcoming LlamaCon AI conference, where further announcements, including the potential release of Llama 4 Behemoth, are anticipated.

Key takeaways:

  • Meta Platforms Inc. has unveiled its latest Llama 4 series of large language AI models, claiming they are among the most powerful open-source models available.
  • The Llama 4 models include Llama 4 Scout and Llama 4 Maverick, with the upcoming Llama 4 Behemoth expected to be the highest performing base model in the world.
  • The models use a "Mixture-of-Experts" system for efficiency and have enhanced safety features to provide balanced responses to sensitive questions.
  • Meta plans to release more AI advancements at its first annual LlamaCon AI conference, potentially including the Llama 4 Behemoth model and a standalone Meta AI application.
View Full Article

Comments (0)

Be the first to comment!