Meta has announced the release of Llama 4, its latest collection of AI models that power Meta AI on the web and in WhatsApp, Messenger, and Instagram Direct. The two new models, Llama 4 Scout and Llama 4 Maverick, are now available for download from Meta or Hugging Face. According to Meta, these models outperform those from OpenAI and Google across a broad range of benchmarks.

Llama 4 Scout is a small model capable of fitting in a single Nvidia H100 GPU, with a 10-million-token context window. Meta claims it beats Google's Gemma 3 and Gemini 2.0 Flash-Lite models, as well as the open-source Mistral 3.1, in various benchmarks. The larger Llama 4 Maverick model is comparable to OpenAI's GPT-4o and Google's Gemini 2.0 Flash, with results similar to DeepSeek-V3 in coding and reasoning tasks using fewer active parameters.

Meta is also training Llama 4 Behemoth, which CEO Mark Zuckerberg claims is "already the highest performing base model in the world." With 288 billion active parameters and 2 trillion parameters in total, Behemoth is expected to outperform its competitors, including GPT-4.5 and Claude Sonnet 3.7, on several STEM benchmarks.

The Llama 4 models utilize a "mixture of experts" (MoE) architecture, which conserves resources by using only the parts of a model that are needed for a given task. This approach allows for more efficient use of computing power and memory. Meta plans to discuss its future plans for AI models and products at LlamaCon, taking place on April 29th.

While Meta calls the Llama 4 collection "open-source," the company has faced criticism for its licenses' less-than-open requirements. For instance, the Llama 4 license requires commercial entities with more than 700 million monthly active users to request a license from Meta before using its models. The Open Source Initiative has stated that this takes the Llama 4 license "out of the category of 'Open Source.'"

The release of Llama 4 marks a significant milestone in the AI model development race, with Meta, OpenAI, and Google continuously pushing the boundaries of what is possible with artificial intelligence. As AI technology advances, it will be interesting to see how these models are used in various applications and industries, and what implications they may have for the future of work and society.

With the AI landscape rapidly evolving, it's essential to stay informed about the latest developments and advancements. As the tech industry continues to innovate, we can expect to see more breakthroughs in AI capabilities and applications. The question remains, however, how these powerful technologies will be used and regulated in the years to come.