Meta Delays Release of “Behemoth” AI Model Amid Performance Concerns
Meta’s long-awaited “Behemoth” AI model will not launch as scheduled. Originally targeted for an April debut, the rollout has now been delayed until autumn or later, according to a Wall Street Journal report citing sources familiar with the situation.
The delay comes amid internal doubts about Behemoth’s performance gains over previous models in the Llama AI family. Engineers have reportedly faced significant training issues, making it difficult to enhance the model's capabilities in a meaningful way. These challenges have raised questions about whether the model is ready for public release.
A Delayed Star at Meta’s Developer Event
Behemoth was intended to headline Meta’s inaugural AI developer conference this past spring. However, with the core model not meeting expectations, Meta instead released Llama 4 Scout and Llama 4 Maverick — two smaller models that offered incremental improvements but didn’t command the spotlight that Behemoth was supposed to own.
Now, the company is under pressure to reassess. Some insiders report growing frustration among Meta’s leadership, and restructuring within the Llama 4 team is being considered. The report suggests that major management changes could be on the table.
Industry-Wide Setbacks in AI Development
Meta isn’t the only company hitting roadblocks. OpenAI’s GPT-5 and Anthropic’s Claude 3.5 Opus have also been delayed, signaling broader headwinds in the generative AI race. Still, Meta claims Behemoth outperforms competitors on select benchmarks, despite internal doubts.
The company may consider launching a limited version of Behemoth in the near future. However, a full-scale release appears unlikely before late 2025.
Meta has yet to confirm a new timeline and declined to comment on the delays. As the AI arms race continues, expectations around cutting-edge performance, safety, and scalability remain high, and so do the risks of falling behind.
