Meta Platforms released its newest Llama 4 AI models on Saturday (5 April) following two delays, stepping up the fight with OpenAI’s ChatGPT and Chinese competitor DeepSeek.

In a statement, Meta touted the multimodal aspect of the Llama 4 Scout and Llama 4 Maverick systems, which can process and integrate various types of data. They are now available for developers to try on Meta’s apps and the Meta.ai website.

Meta added the launch marks “the beginning of a new era for the Llama ecosystem”, as they run as open-weight models, meaning they are capable of working with a range of media beyond text, such as video, images and audio.

The company claims the models are “best in class”, beating rivals including OpenAI’s GPT4o and Gemini 2.0 Flash across “a broad range of widely reported benchmarks”.

However, it teased that the best is yet to come. The Facebook owner said it is previewing the Llama 4 Behemoth, “one of the smartest LLMs in the world and our most powerful yet to serve as a teacher for our new models”.

Delays
Reuters, citing The Information, reported on Friday (4 April) that Meta could end up further delaying the release of Llama 4 due to performance issues, after its launch had already been rescheduled twice.

The news agency explained one of the reasons for the delay is that Llama 4 was not meeting the company’s expectations on technical benchmarks such as reasoning and math tasks.

Meta was apparently also concerned Llama 4 is not as capable as OpenAI’s models when it comes to conducting humanlike voice conversations.

The social media giant has stated it plans to invest $60 billion to $65 billion in 2025 as part of a scheme to broaden its AI infrastructure.

Meta CEO Mark Zuckerberg described Llama 4 as natively multimodal with agentic capabilities during the company’s Q4 2024 earnings call.

“It’s going to be novel, and it’s going to unlock a lot of new use cases,” he said.

*Additional reporting by Mike Robuck