SpaceXAI has announced Grok 4.7, a new version of its flagship model positioned for coding and long-running, agentic work, according to Webrazzi.
The announcement continues a cadence that has compressed model generations into months rather than years. Grok 4.5 entered private beta in July, trained in part on Cursor data, and Grok 4.6 arrived in August with SpaceXAI claiming frontier-class results at roughly half the price of the leading US model. Grok 4.7 lands roughly a month later, and the pitch has shifted from raw benchmark bragging toward the two things buyers of AI capacity actually pay for in 2026: reliable code generation and agents that can hold a task together for hours without being babysat.
That framing is not accidental. The frontier's competitive ground has moved from single-turn chat toward long-horizon work — multi-step software changes, research loops, browser and tool use — where a model's ability to stay coherent across a long context and recover from its own mistakes matters more than any one benchmark point. Positioning a release explicitly around 'coding and long-running tasks' is a direct claim on that market.
Specifics were thin at the time of writing. Pricing, context-window size, benchmark results and the rollout schedule for API and consumer access were not available, and SpaceXAI's own materials are the place to confirm them. The steady drumbeat of releases, though, is itself the story: whatever Grok 4.7 proves to be, it will not be the last model this year.
