Meta unveils a new, more efficient Llama model

Meta a announcement the latest addition to its Llama family of generative AI models: Llama 3.3 70B.

In an article on

“By leveraging the latest advances in post-training techniques…this model improves baseline performance at a significantly lower cost,” Al-Dahle said.

Ahmad Al-Dahle published a chart showing that Llama 3.3 70B outperforms Google's Gemini 1.5 Pro, OpenAI's GPT-4o and Amazon's recently released Nova Pro on a number of industry benchmarks, including MMLU, which evaluates a model's ability to understand and generate text. Via email, a Meta spokesperson said the model is expected to provide improvements in areas such as math, general knowledge, following instructions and using apps.

Llama 3.3 70B, which is available for download from AI development platform Hugging Face and other sources including the official Llama website, is Meta's latest game to dominate the AI ​​field with models “open” that can be used and marketed for a range of purposes.

Meta Terms forced how some developers can use its Llama models; platforms with more than 700 million monthly users must request special permission from the company. But for many developers and businesses, it's irrelevant that Llama models aren't “open” in the strictest sense. According to Meta, its Llama models have accumulated over 650 million downloads.

Meta also exploited Llama for his own purposes. Meta AI, the company's AI assistant powered entirely by Llama models, now has nearly 600 million monthly active users, according to an Instagram account. job by CEO Mark Zuckerberg on Friday. Zuckerberg says Meta AI is poised to become the world's most widely used AI assistant.

Llama's open nature has been a blessing and a curse for Meta.

In November, a report It has emerged that Chinese military researchers have used a llama model to develop a defense chatbot. Meta responded by making his Lama models available to US defense partners.

Meta also expressed concerns about its ability to comply with the AI ​​lawthe European law which establishes a legal and regulatory framework for AI – calling the implementation of the law “too unpredictable”. The problem for the company concerns the related provisions of GDPR, the European privacy law, relating to AI training. Meta trains AI models on public data from Instagram and Facebook users who have not opted out – data which, in Europe, is subject to GDPR safeguards.

EU regulators earlier this year asked Meta to stop training on European user data while they assessed the company's GDPR compliance. Meta gave in, while at the same time approving a open letter calling for “a modern interpretation” of the GDPR that does not “reject progress”.

Meta, not immune to the technical challenges other AI labs face, is scaling up its computational infrastructure to train and serve future generations of llama models. The company announced Wednesday that it will build a $10 billion AI data center in Louisiana – the largest AI data center ever built.

Zuckerberg said during Meta's fourth-quarter earnings call in August, that to train the next major set of Llama models, Llama 4, the company will need 10 times more compute than was needed to train Llama 3.

Training large language models can be expensive. Meta's capital spending increased nearly 33% to $8.5 billion in the second quarter of 2024, from $6.4 billion a year earlier, driven by investments in servers, data centers and network infrastructure.