Meta delays 'Behemoth' AI model, handing OpenAI and Google even more of a head start

Last month, Meta hosted its first generative AI developer conference, LlamaCon, where it planned to release the “Behemoth” large language model. Instead, the company pushed the release to June, and now, a new report suggests it might slip further — to the fall or later.

Also: The top 20 AI tools – and the #1 thing to remember when you use them

According to The Wall Street Journal, Meta engineers are “struggling to significantly improve the capabilities” of Behemoth. That’s especially interesting given that, just weeks ago, Meta claimed Behemoth is still in training and yet already “outperforms” all major competitors — including GPT-4.5, Claude Sonnet 3.7, and Gemini 2.0 Pro — on several STEM benchmarks.

Concerns are brewing

Large language models (LLMs) power AI chatbots such as ChatGPT, Gemini, and Meta AI, processing prompts on the fly and generating human-like text in response. Behemoth was billed as Meta’s “most powerful yet” LLM — “one of the smartest in the world” — and meant to serve as a “teacher” for future models. That’s the public story; internally, concerns are brewing.

Also: Meta’s new AI app delivers a chatbot with a social media twist

Inside Meta, engineers reportedly worry that Behemoth does not offer enough of an improvement over existing models to “justify” a public release. Keep in mind: Meta released Llama 4 in April. (Llama, short for Large Language Model Meta AI, is Meta’s family of LLMs.) To date, two smaller Llama 4 variants, Scout and Maverick, are available, and Meta has teased another lightweight version.

ZDNET requested a comment from Meta and will update this article if a representative responds.

The AI race is real

Meta integrates AI features across Facebook, Instagram, WhatsApp, and Messenger, letting you do things like draft posts and captions, customize chat backgrounds, and even edit images. At the end of April, the company also rolled out a standalone Meta AI app. But the continued delay of Behemoth underscores the challenge Meta still faces in keeping pace with OpenAI and Google.

Also: I test a lot of AI coding tools, and this stunning new OpenAI release just saved me days of work

Ultimately, competitors are shipping new releases at breakneck speed, while Meta still appears to be refining and training an LLM it already announced — and has now repeatedly delayed.

Get the morning’s top stories in your inbox each day with our Tech Today newsletter.