Source: Cointelegraph Original: "{title}"

The White House AI and Crypto Affairs Director David Sacks stated that Meta's latest AI model, Llama 4, has put the United States back in the lead in the global AI race.

“To win the AI competition, the U.S. must also win in the open-source space, and Llama 4 has brought us back to a leading position,” Sacks said in an X post on April 5, as speculation about the U.S.-China competition for global AI dominance continues to heat up.

Since taking office on January 20 under President Donald Trump, Sacks has been outspoken about the AI race. In just over a week in the role, Sacks expressed that he is “confident in America, but we cannot be complacent.”

Sacks' latest comments came after Meta's AI division announced in an X post that it would launch the fourth generation of the Llama model—Llama 4 Scout and Llama 4 Maverick.

Source: David Sacks

“This is our most advanced model to date and the best in its class in the multimodal field,” Meta stated.

Meta claims that its Llama 4 Scout model has 17 billion active parameters and utilizes 16 expert models.

The company asserts that it outperforms competitors' large language models—Gemma 3, Gemini 2.0 Flash-lite, and Mistral 3.1—in “widely recognized multiple benchmark tests.”

Meanwhile, Llama 4 Maverick also has 17 billion active parameters but is configured with 128 expert models. Meta claims that its Maverick model can surpass GPT-4o and Gemini 2.0 Flash in “widely recognized multiple benchmark tests.”

Llama 4 Maverick instruction tuning benchmark tests. Source: Meta

Meta also stated that Maverick's performance on “reasoning and coding tasks” is comparable to DeepSeek v3, despite using only half the active parameters.

Less than a year ago, in July 2024, Meta CEO Mark Zuckerberg stated that he expects the Llama model to become “the industry's most advanced model” by 2025. It has been over two years since Meta first released a limited version of Llama 1 in February 2023.

At that time, Meta expressed that it was “shocked” by the demand, receiving over 100,000 access requests.

Related: Perceptive open-source AI search outperforms GPT-4o and Perplexity

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。