Last month, Meta was the first ever generated AI conference. But while the campaign has brought some significant improvements to developers, it is also a bit overwhelmed considering the importance of AI to the company. Now, we know more about why, thanks to The Wall Street Journal.
According to the report, Meta initially intended to release its “behemoth” Llama 4 model at an April developer event, but later postponed its release until June. Now, obviously being pushed back again, it is possible until “fall or later.” Metaenges are reportedly “working to greatly improve the capabilities of what Mark Zuckerberg calls the “the world’s highest performing fundamental model.”
Meta has released two smaller Llama 4 models and has also mocked the fourth lightweight model, apparently nicknamed “Little Llama”. Meanwhile, the Behemoth model will have 288 billion active parameters and “better than several STEM benchmarks on GPT-4.5, Claude Sonnet 3.7 and Gemini 2.0 Pro,” the company said.
Meta never gave a solid timeline for when to look for the model. The company said last month that it was “still training.” Although the behemoth gets some nods in the Cougar keynote, there is no latest news yet when it can be really ready. That might be because it could be a few months. In meta, there is clearly a question: “Ask for whether improvements to previous versions are sufficient to justify public releases.”
Meta did not immediately respond to a request for comment. As the report points out, it is not the first company to face obstacles as it will release new models and surpass competitors. However, given that meta involves AI, latency is still worth noting. Zuckerberg’s Meta-Man AI plans to spend up to excess on its AI infrastructure this year.
