Meta Releases Llama 3.1, It’s Most Powerful Open-Source A.I. Model Yet

Mark Zuckerberg says he expects Llama A.I. models to become the most advanced in the industry by 2025.

As Big Tech’s A.I. race heats up, Meta is striking out on its own path. The social media giant yesterday (July 23) released Llama 3.1 405B, which it says is the largest-ever open-source A.I. model to rival the capabilities of competitors like OpenAI’s GPT-4o and Anthropic’s Claude 3.5 Sonnet. Unlike closed A.I. systems, which are favored by major tech players like OpenAI, Anthropic and Google, open-source A.I. models can be used and downloaded at no cost. Meta has long emphasized an open-source approach toward its projects, claiming such technology should be shared. “Open source will ensure that more people around the world have access to the benefits and opportunities of A.I., that power isn’t concentrated in the hands of a small number of companies, and that the technology can be deployed more evenly and safely across society,” Meta CEO Mark Zuckerberg wrote in a blog post announcing Llama 3.1 yesterday.

Llama 3.1 405B will power Meta AI, Meta’s chatbot featured across its apps. The model has enhanced reasoning capabilities, support for several new languages, and features like an “Imagine me” prompt that generates customized images. Trained with more than 16,000 Nvidia (NVDA) graphics processing units (GPUs) and boasting 405 billion parameters, Llama 3.1 405B is Meta’s largest A.I. model thus far and was released alongside two smaller versions of the Llama 3.1 model.

While Meta has released open-source A.I. models in the past, it is positioning Llama 3.1 405B as the first to truly match rival products. According to the company, Llama 3.1 405B displays superior capabilities across numerous long-context benchmarks compared to GPT-4o and Claude 3.5 Sonnet. It also said its model contains higher capabilities in math, reasoning and tool use.

“Today, several tech companies are developing leading closed models. But open source is quickly closing the gap,” Zuckerberg said in the blog post, adding that he expects future Llama models to become the most advanced in the industry by 2025. With its open-source approach, Meta hopes its A.I. model will become the foundation of other A.I. ventures pursued by companies and startups across the industry. It plans to work with companies like Nvidia, Amazon (AMZN) and Databricks to help developers as they fine-tune and distill new versions of Llama 3.1

Is open source A.I. the way forward?

Some of the most significant criticisms against open-source A.I. concern bad actors’ potential to use the technology to create harm. While Zuckerberg conceded that this may occur in the future, he noted that open-source software can also help safeguard against such risks. “I think it will be better to live in a world where A.I. is widely deployed so that larger actors can check the power of smaller bad actors,” he said. “As long as everyone has access to similar generations of models—which open source promotes—then governments and institutions with more compute resources will be able to check bad actors with less compute.”

Meta’s release and Zuckerberg’s cry for open-source A.I. drew praise from leaders across the tech industry. “This is the way,” tweeted Twitter founder Jack Dorsey. Aravind Srinivas, CEO of A.I. startup Perplexity, described Zuckerberg’s open-source argument as “well articulated,” while Andrej Karpathy, a former OpenAI and Tesla (TSLA) researcher, congratulated Meta on the launch. Even Elon Musk, who has long clashed with Zuckerberg, described the initiative as “impressive” and noted that Zuckerberg “does deserve credit for open-sourcing.”