Last month, Meta hosted LlamaCon, its first ever generative AI conference. But while the event delivered some notable improvements for developers, it also felt a bit underwhelming considering how important AI is to the company. Now, we know a bit more about why, thanks to a new report in The Wall Street Journal.
According to the report, Meta had originally intended to release its “Behemoth” Llama 4 model at the April developer event, but later delayed its release to June. Now, it’s apparently been pushed back again, potentially until “fall or later.” Meta engineers are reportedly “struggling to significantly improve the capabilities” of the model that Mark Zuckerberg has called “the highest performing base model in the world.”
Meta has already released two smaller Llama 4 models, Scout and Maverick, and has also teased a fourth lightweight model that’s apparently nicknamed “Little Llama.” Meanwhile, the “Behemoth” model will have 288 billion active parameters and “outperforms GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro on several STEM benchmarks,” the company said last month.
Meta has never given a firm timeline of when to expect the model. The company said last month that it was “still training.” And while Behemoth got a few nods during the LlamaCon keynote, there were no updates on when it might actually be ready. That’s probably because it could still be several months. Inside Meta there are apparently questions “about whether improvements over prior versions are significant enough to justify public release.”
Meta didn’t immediately respond to a request for comment. As the report notes, it wouldn’t be the first company to run into snags as it races to release new models and outpace competitors. But the delay is still notable given the Meta’s lofty ambitions when it comes to AI. Zuckerberg has made AI a top priority with Meta planning to spend as much as $72 billion on its AI infrastructure this year.
