Meta Platforms has decided to delay the release of its highly anticipated AI model, “Behemoth” due to concerns about its performance and whether it offers significant improvements over previous versions. Originally planned for an April launch to coincide with Meta’s first AI developer conference, LlamaCon, the release was first pushed to June and now is expected in the fall or possibly later.
Company engineers have encountered difficulties enhancing Behemoth’s capabilities enough to justify a public debut, leading to internal debates about the model’s readiness. This has also caused frustration among senior executives, with some reportedly blaming the Llama 4 development team for the setbacks. Meta is even considering major management changes within the AI product group responsible for Behemoth.
Other articles you may find interesting
Despite the delay, Meta recently introduced smaller versions of its Llama 4 large language model, namely Llama 4 Scout and Llama 4 Maverick, which have been well received. The company had touted Behemoth as one of the smartest and most powerful large language models globally, intended to serve as a “teacher” for future AI models.
This postponement comes amid Meta’s massive investment in AI, with plans to spend up to $72 billion on capital expenditures this year, mostly focused on AI development. The delay has raised questions about the broader AI industry’s reliance on simply scaling up model size as a path to progress, a challenge also faced by other AI leaders like OpenAI and Google, who have similarly encountered delays with their next-generation models.
Meta’s AI journey began with its Fundamental AI Research Team, creators of the original Llama models in early 2023. However, many key researchers have since left, and newer iterations, including Behemoth, are being developed by a different team. This transition could be contributing to the current development challenges.
In summary, Meta’s delay of the Behemoth AI model signals a cautious approach amid technical hurdles and internal scrutiny, reflecting broader industry challenges in advancing large language models beyond current capabilities. The company may still consider releasing a limited version earlier, but no new public timeline has been announced by CEO Mark Zuckerberg.
