Meta Nears Release of New AI Model After Performance Hiccups
The company’s Llama 4 model is running behind schedule at a moment when Meta is scrambling to get to the front of the pack in AI.
Last spring, Meta Platforms was on a tear. It had just released Llama 3, a new generation of its large language model, to much praise from developers and independent reviewers. Almost a year later, though, that model’s successor, Llama 4, is running behind schedule.
Meta has pushed back the release date of Llama 4 at least twice and could do so again, according to two people familiar with the matter. Currently, it’s aiming to release Llama 4 later this month, they said.
One reason for the delay: During its development, Llama 4 didn’t perform as well on technical benchmarks as Meta had hoped, particularly on reasoning and math tasks, the two people said. Meta has also had concerns about Llama 4 being less capable than OpenAI’s models at conducting humanlike voice conversations, they said.