Meta Unveils Llama 4: A Major Leap in Open-Source AI Development

ByTech NEWS
Meta Unveils Llama 4

Meta has officially introduced Llama 4, the latest addition to its open-source AI model lineup. This cutting-edge model series was announced on April 5. It includes Scout, Maverick, and the upcoming Behemoth.

All are designed to elevate AI performance across platforms. With Llama 4, Meta takes a bold step. The company joins the ongoing AI race alongside industry leaders such as Google and OpenAI.

A New Generation of AI Models

In a video posted to Instagram, Meta CEO Mark Zuckerberg reaffirmed the company’s commitment to building world-class open-source AI. “Our goal is to build the world’s leading AI, open-source it, and make it universally accessible,” he said. Llama 4 is a clear manifestation of that vision.

Two models—Llama 4 Scout and Llama 4 Maverick—are now available for download through Meta’s official Llama website and Hugging Face. These models also power Meta AI, the digital assistant integrated into platforms such as WhatsApp, Instagram, Messenger, and more.

What Makes Llama 4 Stand Out?

The Llama 4 series is Meta’s first to adopt a mixture-of-experts (MoE) architecture. This approach divides a large language model into smaller, specialized components or “experts.”

These experts focus on specific domains like biology, physics, literature, or programming. Only the relevant experts activate for a given task. The result is a more efficient, high-performance model—both in training and execution.

Llama 4 Scout:

  • 17 billion parameters
  • 16 experts
  • 10 million-token context window
  • Designed to run on a single GPU, ideal for lightweight tasks

Llama 4 Maverick:

  • Also with 17 billion parameters
  • Uses 128 experts
  • Designed for general assistant and chatbot applications
  • Optimized for single-host deployment

Meta claims that Maverick not only surpasses GPT-4o and Gemini 2.0 Flash in various benchmarks. These benchmarks include coding, reasoning, image understanding, and long-context processing. Maverick also performs comparably to the much larger DeepSeek v3.1.

Behemoth and What’s Coming Next

Currently in development, Llama 4 Behemoth is set to become one of the most powerful models in Meta’s AI portfolio. It is expected to feature:

  • 288 billion active parameters
  • 16 experts
  • Nearly 2 trillion total parameters

According to Meta, Behemoth surpasses the performance of GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro in multiple STEM-related evaluations.

Zuckerberg also hinted at the development of Llama 4 Reasoning. It is a specialized model focused on complex problem-solving. The model is also designed for general reasoning capabilities. More details are expected in the near future.

Rapid Growth and Heavy Investment

Llama’s popularity continues to surge. Meta reported that the model suite exceeded one billion downloads. This occurred just two weeks before the launch of Llama 4. It was up from 650 million in December 2024.

To support its growing AI initiatives, Meta has outlined plans for significant capital investment. The company plans to allocate between $60 billion and $65 billion in 2025. This allocation is for building infrastructure like servers and data centers dedicated to AI development.

Final Thoughts

The release of Llama 4 marks a major milestone in Meta’s AI strategy. Meta emphasizes openness, scalability, and high performance. By doing so, Meta positions itself as a serious contender in the next era of generative AI.

More models are in the pipeline. Massive infrastructure investment is also on the horizon. Llama 4 is only the beginning of Meta’s next AI chapter.

Related Posts