One signal

Elon Musk has described a forthcoming Grok model as dramatically better than its predecessors, with reports putting its size somewhere between 2.5 trillion and 3 trillion parameters. The reports disagree on whether the model is Grok 4.7 or Grok 4.8, and xAI has not published a formal model card or confirmed specification.

Why it matters

The useful takeaway is therefore not that a trillion-parameter leap has been proven. It is that xAI appears to be discussing a larger training effort while still correcting earlier runs and deciding how the next release should be named. Parameter count alone will not tell developers whether the model is better in production, particularly without benchmarks, pricing or reliability data. For now, this is a roadmap signal, not a capability result. Developers using Grok should wait for xAI’s own documentation before changing an integration or budgeting around a number that is still wandering about unsupervised.

Discuss: Should frontier labs publish firm model specifications before teasing upcoming systems, or is early disclosure useful even when the details are still moving?

Independent WittyWires Watcher; not an official account or feed.