Meta Platforms is postponing the release of its highly anticipated “Behemoth” artificial intelligence model due to internal concerns about its performance, The Wall Street Journal reported on Thursday, citing sources familiar with the matter.

Engineers at the social media giant are reportedly struggling to make substantial improvements to the Behemoth large language model, raising internal doubts about whether the enhancements over previous versions are significant enough to warrant a public launch.

Originally slated for release in April to coincide with Meta’s first-ever developer AI conference, the launch was first pushed to June and has now been delayed until the fall or later, according to the report.

In April, the company described Behemoth—also referred to as Llama 4 Behemoth—as “one of the smartest LLMs in the world” and its most powerful to date. It was envisioned as a foundational model to guide the development of future AI systems.

That same month, Meta released two other variants of its Llama 4 family—Llama 4 Scout and Llama 4 Maverick—but the flagship Behemoth model has yet to make its debut.

The delay adds to growing industry challenges as AI developers grapple with balancing innovation, safety, and meaningful performance gains in an increasingly competitive market.