xAI targets September 12 launch for Grok 4.7 with 2.1T parameters

Grok 4.7's headline claim is scale plus proprietary data: a jump to roughly 2.1 trillion parameters, up 40% from Grok 4.6, paired with a supplemental training phase drawing on SpaceX engineering records and Starlink satellite telemetry. That data provenance is the differentiator xAI is leaning on — the argument being that access to real-world aerospace and networking data gives Grok domain grounding rivals can't easily replicate. Elon Musk claims across-the-board gains over the prior version with improved token efficiency.

The September 12 date makes this a near-term, concrete event rather than vaporware, but official documentation is not yet published, so parameter counts and efficiency claims remain company-sourced. That matters given how the week's other releases arrived with third-party benchmarks in hand.

Competitively, xAI has been positioning Grok as a cost-and-coding play — recent Grok 4.5/4.6 releases at roughly $2/$6 per million tokens claimed steep cost cuts while independent reporting noted the models still trailed OpenAI and Anthropic on capability. Grok 4.7 is the attempt to close that capability gap while keeping the pricing edge. The skeptical read: parameter count alone doesn't guarantee frontier performance, and a 40% size bump raises inference-cost questions that cut against the affordability narrative. Whether the SpaceX/Starlink data translates into measurable general-purpose gains — versus narrow aerospace competence — is the thing to watch when benchmarks land on the 12th.