According to axios.com, the release of Meta’s Llama 4 Behemoth model will be delayed until fall 2025 or later. Apparently, folks at Meta are concerned that it might not be a sufficient advancement over the competition.

I can’t imaging training and testing a model with nearly two trillion total parameters. I hope Meta’s engineers take all the time they need to make the model as good as it can be. I doubt I’ll ever have the computing power to run the model, but I can certainly dream…