AIBRIEF
Grok 4.7 would jump from 4.6's 1.5T parameters and add SpaceX data training; Musk's 'exceed all current models' claim waits on independent benchmarks.
Elon Musk's AI company xAI says its next Grok model will release on Sept. 12 with 2.1 trillion parameters, roughly 40% larger than the 1.5T-parameter Grok 4.6 released Aug. 12. The announcement came Tuesday in an X post from Musk, and was restated by NextBigFuture the same day.
Two concrete deltas are worth testing on release day. First, the parameter jump from 1.5T to 2.1T, which Musk says will make Grok 4.7 "better than 4.6 in every way except slightly slower to serve." Second, supplemental training on a "large amount of SpaceX company data," a notable provenance choice, since xAI and SpaceX are separate companies, and one that raises questions about how the corporate dataset was assembled and licensed.
Musk's bigger claims, that Grok 4.7 will "exceed all current models" and be the "best at real-world engineering," are stakeholder self-claims. No independent benchmark in the receipts confirms them. The Artificial Analysis baseline for Grok 4.6 gives a starting point, but the test on Sept. 12 is whether the new model moves on public evals like GDPval, HLE, and GPQA Diamond.
xAI's release notes remain the canonical place to confirm the actual GA, parameter count, and any disclosed SpaceX-data details. The model has not shipped yet.