Meta unveil two Llama 4 models (and one beast)

Meta unveil two Llama 4 models (and one beast)

Summary: Meta officially released its new Llama 4 AI models. One is lightweight (able to fit on a single GPU), while the other is much larger. Meta plans to take on the best AI models on the market.

When we speak about the best AI models, we typically talk about offerings from OpenAI, Google, and DeepSeek. However, Meta is also one of the frontrunners in the AI market, and its models are pretty powerful.

On Monday, Meta launched two Llama 4 models, and the company believes that it has the other AI giants cornered. These models will be used to power the AI in platforms like Instagram, Facebook Messenger, and WhatsApp

Meta launches its new Llama 4 models

Just like other AI companies, Meta has a collection of models of varying sizes and capabilities. The first model is called Llama 4 Scout, and it’s the smaller of the pair. This one has 17 billion parameters and 16 experts. All in all, it has 109 billion active parameters.

What’s mind-boggling about this model is its context window. Do you remember how impressive it was when Gemini introduced the 1 million-token context window? Well, Llama 4 Scout has a massive 10 million-token context window. This puts it on a new level of memory.

Even with such a massive context window, Llama 4 Scout is small enough to fit on a single Nvidia H100 GPU. It’s better fitted for personal use and more moderate tasks.

Maverick

Llama 4 Scout has a bigger brother named Llama 4 Maverick. This model is more of an industrial offering aimed at taking on the likes of Gemini 2.5 Pro, OpenAI’s 4o, and DeepSeek’s V3 models. It’s not exactly class-leading, but it’s enough for business use and larger projects.

This model has the same 17 billion parameters, but it has 128 experts. This gives it 400 billion total parameters. These specs are more robust than what you get with Scout, but there’s one exception. Llama 4 Maverick comes with a 1 million-token context window.

These are two extremely powerful models, and they’re available to download now. But, that’s not all that Meta has to offer, as it’s hiding a “behemoth”.

Behemoth

Llama 4 Behemoth will be Meta’s top-tier model to take on the best that the other companies have to offer. This one will have 288 parameters with 16 experts. It will come with a scale-breaking 2 trillion active parameters.

Llama 4 Behemoth will be Meta’s most impressive model when it launches. Right now, it’s in preview, and we don’t know when Meta will officially launch it. However, we’ll find out more during LlamaCon on April 29th.

📰 Crime Today News is proudly sponsored by DRYFRUIT & CO – A Brand by eFabby Global LLC

Design & Developed by Yes Mom Hosting

Crime Today News

Crime Today News is Hyderabad’s most trusted source for crime reports, political updates, and investigative journalism. We provide accurate, unbiased, and real-time news to keep you informed.

Related Posts