Meta Adds 'Multimodal' Models to Its Llama AI Stable
Meta has released the latest versions of its Llama artificial intelligence (AI) model. The tech giant says its Llama 4 models, unveiled Friday (April 4), are built on one of the most advanced large language models (LLMs) in the world.
New Offerings
Among the new offerings are Llama 4 Scout and Llama 4 Maverick, which Meta calls “the first open-weight natively multimodal models,” multimodal meaning able to work with media other than text. “We’re also previewing 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,” the announcement said.
AI Investment
Meta has been investing heavily in AI over the last two years, with chief executive Mark Zuckerberg announcing plans to spend up to $65 billion in 2025 to strengthen the company’s artificial intelligence stable. As noted here last month, the company is looking to extend its AI capabilities beyond its social media businesses and is mulling trials of premium subscriptions for its AI assistant, Meta AI, for agentic purposes such as booking reservations and video creation.
OpenAI's Move
Also on Friday, PYMNTS wrote about plans by OpenAI to release an open-source version of its LLM, something the company has not done in years. An open-source model is one that would be generally free to use, modify, and distribute. The creator of ChatGPT revealed the news through a form seeking feedback for this strategy, saying it wants to hear from developers, researchers, and the public to help it “make this model as useful as possible.” The open-source model would be coming “in months,” the company said.
Conclusion
“OpenAI’s decision comes as open-source models like Meta’s Llama, Mistral’s LLM, and DeepSeek have been gaining in popularity,” PYMNTS wrote. “In March, Meta CEO Mark Zuckerberg said on Threads that Llama has been downloaded 1 billion times. Llama was launched in 2023.”
