Meta has introduced a new collection of AI models named Llama 4, expanding its Llama family. This set includes four models: Llama 4 Scout, Llama 4 Maverick, and Llama 4 Behemoth, all trained on extensive amounts of unlabeled text, images, and video data to enhance their "broad visual understanding."
Following the success of open models from Chinese AI lab DeepSeek, which reportedly performed better than Meta's previous models, Meta accelerated the development of Llama. The new models have been made available on Llama.com and through partners like Hugging Face, while the Behemoth model is still under development. However, there are restrictions for users from the EU due to regional AI and data privacy laws.
Maverick features 400 billion total parameters with 17 billion active parameters and has outperformed models like OpenAI's GPT-4 in specific areas. Meanwhile, Scout excels in tasks such as document summarization and can handle up to 10 million tokens. Meta claims these new models are designed to provide more inclusive responses to contentious topics, aiming to answer a wider range of questions without bias.
