On Saturday, Meta released the latest iteration of its large language model (LLM), named Llama 4 Scout and Llama 4 Maverick. Meta describes Llama as a multimodal AI system, meaning it can process and integrate diverse data types, including text, video, images, and audio, and convert content across these formats. In a statement, Meta asserted that Llama 4 Scout and Llama 4 Maverick are its “most advanced models yet” and “the best in their class for multimodality.
Furthermore, Meta announced that Llama 4 Maverick and Llama 4 Scout will be open-source software. The company also previewed Llama 4 Behemoth, which it characterized as “one of the smartest LLMs in the world and our most powerful yet,” intended to serve as a teacher for its new models.
This development occurs amidst aggressive investments in artificial intelligence (AI) infrastructure by major technology firms, spurred by the success of OpenAI’s ChatGPT, which significantly reshaped the tech landscape and fueled investment in machine learning.
Prior to this release, The Information reported on Friday that Meta had delayed the launch of this LLM version. This delay stemmed from Llama 4’s initial performance falling short of Meta’s technical benchmark expectations, particularly in reasoning and mathematical tasks. Concerns also arose regarding Llama 4’s comparative capability in conducting human-like voice conversations relative to OpenAI’s models, according to the report.
Meta plans to allocate up to $65 billion this year to expand its AI infrastructure, responding to investor pressure on big tech firms to demonstrate returns on their AI investments.