The latest version of Meta Platforms’ large language model (LLM) Llama, known as Llama 4 Scout and Llama 4 Maverick, was released on Saturday, April 5, as the company tries to take the lead in generative AI investments.

David Sacks, the White House’s newly appointed AI and crypto czar, has praised the tech firm’s latest open-source AI model, Llama 4, as a major leap forward for the United States in the international race for AI dominance.

“For the US to win the AI race, we have to win in open source too, and Llama 4 puts us back in the lead,” Sacks said in an April 5 X post, as speculation continues to mount over the US and China competing for the top spot in the global AI race.

Since assuming his role following President Donald Trump’s January 20 inauguration, Sacks has remained outspoken on the need for urgency and innovation. Just over a week into the job, Sacks said he is “confident in the US, but we can’t be complacent.”

Meta unleashes Llama 4 models to lead multimodal AI race

Llama is a multimodal artificial intelligence system, according to Meta. Multimodal systems can process and integrate different kinds of data, such as audio, video, images, and text, and are capable of converting content across these formats.

The new models now include the Llama 4 Scout, Llama 4 Maverick, and Llama 4 Behemoth. 

Following the success of OpenAI’s ChatGPT, which changed the tech landscape and sparked investment in machine learning, giant technology companies have aggressively invested in artificial intelligence (AI) infrastructure.

In an announcement, Meta claimed that the Llama 4 Scout and Llama 4 Maverick were its “most advanced models yet” and “the best in their class for multimodality.” 

Moreover, the company also disclosed that Llama 4 Maverick and Llama 4 Scout would also be released as open-source software.

Meta also announced that it was previewing Llama 4 Behemoth, which it called “our most powerful yet to serve as a teacher for our new models and one of the smartest LLMs in the world.”

According to reports, the latest version of the model was held back during development as Llama 4 did not meet Meta’s technical benchmarks, particularly concerning math and reasoning tasks. Additionally, it was feared that Llama 4 would not stack up with OpenAI’s models regarding human-like voice conversations.

Meta plans to spend up to $65 billion this year on building its AI infrastructure as investors push the biggest tech companies to deliver a return on their capital.

Meta invested in Llama 4 Maverick and Scout due to their reputable attributes 

The two models of the most recent version of Llama 4, Maverick and Scout, possess unique qualities that make them better suited for a given task.

Starting with Llama 4 Maverick, Meta’s internal testing indicates that Maverick outperforms models like OpenAI’s GPT-4o and Google’s Gemini 2.0 on specific coding, reasoning, multilingual, long-context, and image benchmarks. 

Moreover, the company claims that Maverick is best for “general assistant and chat” use cases like creative writing. Maverick, however, falls short of more powerful newer models such as OpenAI’s GPT-4.5, Anthropic’s Claude 3.7 Sonnet, and Google’s Gemini 2.5 Pro.

Additionally, Llama 4 Scout has also proven to have incredible features. Scout excels in tasks like document summarization and reasoning over sizable codebases. 

Furthermore, Scout can process and work with very long documents because it can take in images and up to millions of words in plain English. One of its unique features is its enormous context window of 10 million tokens.

Meta expands Llama 4 to 40 countries via Meta AI

Meta claims to have updated Meta AI, its AI-powered assistant that works with apps like Instagram, WhatsApp, and Messenger, to use Llama 4 in 40 countries. Currently, multimodal features are only available in English in the U.S.

Notably, Scout and Maverick are freely accessible on Llama.com. However, users or businesses that are “domiciled” or have a “principal place of business” in the EU cannot use or distribute the models. This is probably due to the laws about AI and data privacy being enforced as governance requirements in the region, which Meta previously referred to as too difficult.

Like earlier Llama releases, businesses with over 700 million monthly active users must apply for a special license from Meta, which the company may or may not grant. Additionally, Meta stated that some developers might disagree with the Llama 4 license.

Based on Meta’s blog post, “the Llama ecosystem is entering a new era with these Llama 4 models. This is only the start of the Llama 4 collection.”

Cryptopolitan Academy: Coming Soon - A New Way to Earn Passive Income with DeFi in 2025. Learn More