Meta, Yeni “Llama 3.3 70B” Modelini Tanıttı

Meta, which owns Facebook, Instagram, and WhatsApp, is continuously enhancing its Llama artificial intelligence models. The latest addition is Llama 3.3 70B, which has arrived.

In a statement about the new model, it was noted: “Llama 3.3 70B offers the performance of our 405B model, yet it is our newest model that is easier to run and more cost-effective. Utilizing the latest advancements in post-training techniques, including online optimizations, the 70B significantly enhances performance at a lower cost, making it a more accessible option for the entire open-source community.” Previously, the opening of Llama models to government agencies and companies operating in national security sectors had garnered significant attention. Numerous private sector firms and government agencies—partnered with the U.S. government, including major defense companies like Palantir and Lockheed Martin, as well as Amazon Web Services, Oracle, and Microsoft—will now have seamless access to Meta’s Llama AI models. It is expected that this move will accelerate ongoing efforts and deliver substantial benefits across various fields. For instance, Oracle uses Llama to synthesize aircraft maintenance documents, enabling technicians to diagnose issues more quickly and accurately, thereby speeding up repair times and bringing critical aircraft back into service.

Reports indicate that Meta has also provided similar access to Llama for governments and private firms collaborating with the state in countries like the UK, Canada, Australia, and New Zealand. Meta’s CEO, Mark Zuckerberg, during the last quarterly discussions, mentioned that the company’s new open-source LLM, called Llama 4, has been trained on a cluster of over 100,000 Nvidia H100 AI GPUs—the largest cluster of its kind compared to what others have developed. The company has invested billions of dollars in server infrastructure to bolster AI systems, with plans to launch Llama 4 in the first quarter of 2025.