Meta Platforms unveiled the latest iteration of its large language model (LLM), Llama, on Saturday.
The company introduced two new versions, Llama 4 Scout and Llama 4 Maverick, calling them its “most advanced models yet” and positioning them as leaders in multimodal AI technology. These models are designed to process and integrate diverse data types, including text, images, video, and audio, allowing them to convert content across various formats.
Meta also announced that both Llama 4 Scout and Llama 4 Maverick would be open-source software, further enhancing their accessibility for developers. Additionally, Meta previewed Llama 4 Behemoth, describing it as one of the most powerful LLMs in the world, intended to serve as a teacher for the newer models.
The launch of these models follows a surge of investment in AI infrastructure by major technology companies, spurred by the success of OpenAI’s ChatGPT. However, Meta had initially delayed the launch of Llama 4 after it failed to meet internal technical benchmarks, particularly in reasoning and math tasks.
There were also concerns that Llama 4 could not match OpenAI’s capabilities in humanlike voice conversations.
To support its AI efforts, Meta plans to invest up to $65 billion this year in expanding its AI infrastructure, responding to increasing investor demands for returns on these investments.
