Meta Nears Launch of New AI Mannequin Llama 4 in April: Report

Meta Platforms plans to launch the newest model of its massive language mannequin later this month, after delaying it a minimum of twice, the Info reported on Friday, because the Fb proprietor scrambles to steer within the AI race.

Meta, nonetheless, may push again the discharge of Llama 4 once more, the report stated, citing two folks acquainted with the matter.

Massive expertise companies have been investing aggressively in AI infrastructure following the success of OpenAI’s ChatGPT, which altered the tech panorama and drove funding into machine studying.

The report stated one of many causes for the delay is throughout growth, Llama 4 didn’t meet Meta’s expectations on technical benchmarks, significantly in reasoning and math duties.

The corporate was additionally involved that Llama 4 was much less succesful than OpenAI’s fashions in conducting humanlike voice conversations, the report added.

Meta plans to spend as a lot as $65 billion (roughly Rs. 5,39,000 crore) this 12 months to develop its AI infrastructure, amid investor strain on large tech companies to point out returns on their investments.

Moreover, the rise of the favored, lower-cost mannequin from Chinese language tech agency DeepSeek challenges the idea that creating the perfect AI mannequin requires billions of {dollars}.

The report stated Llama 4 is predicted to borrow sure technical features from DeepSeek, with a minimum of one model slated to make use of a machine-learning approach known as combination of specialists methodology, which trains separate elements of fashions for particular duties, making them specialists in these areas.

Meta has additionally thought-about releasing Llama 4 by Meta AI first after which as open-source software program later, the report stated.

Final 12 months, Meta launched its principally free Llama 3 AI mannequin, which may converse in eight languages, write higher-quality pc code and clear up extra advanced math issues than earlier variations.

© Thomson Reuters 2025