Meta Launches Llama 4: Everything You Need to Know About the New Open-Source AI Models

In a major leap toward dominating the generative AI race, Meta released the first models from its latest AI suite, Llama 4, on Saturday. With this launch, the tech giant aimed to strengthen its position in the global AI market and challenge competitors like OpenAI and Google.

Meta CEO Mark Zuckerberg announced the release through an Instagram video, where he laid out the company’s long-term vision. “Our goal is to build the world's leading AI, open source it, and make it universally accessible... I've said for a while that open-source AI will lead the way, and with Llama 4, we're starting to see that happen,” he said.

The first two models in the Llama 4 family — Llama 4 Scout and Llama 4 Maverick — were made available for download through Meta’s official Llama website and open-source platform Hugging Face. Both models also power Meta AI, the company’s virtual assistant now embedded across WhatsApp, Instagram, Messenger, and the web.

Meta Introduces ‘Behemoth’ and Mixture-of-Experts Model for Smarter, Efficient AI

Along with Scout and Maverick, Meta introduced Llama 4 Behemoth, its most powerful model yet, which remains under development. Behemoth has been described as one of the smartest LLMs (large language models) created so far and is expected to shape the future training of Meta’s AI systems.

With this release, Meta used the mixture-of-experts (MoE) framework for the first time. MoE divides the AI model into specialized units or “experts,” each trained in different areas like physics, poetry, biology, and programming. For every task, only the most relevant experts are activated. This technique helps reduce energy and training costs while improving efficiency.

Also Read: 9 Nations, 44 Naval Personnel: Rajnath Singh Flags off INS Sunayna from Karnataka's Karwar

Llama 4 Scout and Maverick

Llama 4 Scout was built with 17 billion parameters and 16 experts, and it offers a context window of 10 million tokens. It is designed to run on a single GPU, making it highly efficient for lightweight deployment. This puts it in the same category as Google’s recently launched Gemma 3, which also focuses on small, high-performance models.

Llama 4 Maverick, also with 17 billion parameters but equipped with 128 experts, was developed for a wider range of assistant-style applications. Meta called it a general-purpose “workhorse” for digital tasks like chat, reasoning, and problem-solving.

According to Meta, Maverick outperformed OpenAI’s GPT-4o and Google’s Gemini 2.0 Flash in multiple benchmarks such as code generation, image analysis, multilingual reasoning, and long-context handling. It also rivaled DeepSeek v3.1, despite DeepSeek being a much larger model.

Llama 4 Behemoth and Reasoning Model Coming Soon

While Llama 4 Behemoth is still being developed, Meta revealed that it is expected to feature 288 billion active parameters and nearly 2 trillion total parameters. In internal tests, it reportedly outperformed GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro in STEM-related benchmarks.

Zuckerberg also hinted at the upcoming launch of Llama 4 Reasoning, a new model tailored specifically for complex problem-solving and analytical use cases. Meta promised more updates on this variant in the coming weeks.

“This is just the start of the Llama 4 lineup,” Meta wrote in its official blog. “We believe the most advanced AI systems must be capable of taking generalized actions, engaging in natural conversations, and tackling problems they've never encountered before.”

Meta’s AI Expansion

Meta stated that Llama models had been downloaded more than one billion times—a significant increase from 650 million downloads reported in December 2024. This growth demonstrated the rising popularity of Meta’s open-source AI tools among researchers and developers worldwide.

In January, Zuckerberg revealed that Meta would invest between $60 billion and $65 billion in 2025 to scale up its AI infrastructure. The money would be used for servers, data centres, and other hardware to support the growing AI ecosystem within the company.