What will be true of Grok 3's parameter counts?

xAI recently open-weighted Grok 2, revealing that it has 115B active parameters and 270B total parameters (per internet claims in various places that I have not vetted closely). xAI also plans to open-weight Grok 3 in about six months.

Grok 3 was trained in a data center containing 100k+ H100 GPUs; xAI claims that Grok 3 was pre-trained on more than 10x as much FLOP as Grok 2 (training FLOP scales in proportion to active parameters, as well as the size of the training data).

N/A if Grok 3 is not opened in one year, and we never get confirmation or a credible report of Grok 3's params by that date.