Meta has just unveiled a new addition to its Llama series of generative AI models: Llama 3.3 70B. According to Ahmad Al-Dahle, Meta’s VP of generative AI, this latest model offers similar performance to the larger Llama 3.1 405B, but at a significantly lower cost. By utilizing advanced post-training techniques, including online preference optimization, Llama 3.3 70B is more cost-efficient while maintaining high core performance.
In benchmark tests, Llama 3.3 70B outperforms several notable competitors in the AI space, such as Google’s Gemini 1.5 Pro, OpenAI’s GPT-4, and Amazon’s Nova Pro. These comparisons were made on industry-standard evaluations like MMLU, which tests the model’s ability to understand and process language. According to Meta, Llama 3.3 70B shows notable improvements in areas like mathematics, general knowledge, instruction following, and app usage.
Meta is positioning Llama 3.3 70B as a competitive force in the generative AI field with its “open” models. While there are usage restrictions for larger platforms (those with more than 700 million monthly users need special licenses), Llama models have already seen significant adoption, with over 650 million downloads. The model is available for download on platforms like Hugging Face and the official Llama website.
With this release, Meta is further cementing its position in the AI landscape, offering powerful tools that can be used and commercialized for various applications