xAI Ships Grok 4.7 at 2.1T Parameters, Still Chasing the Frontier 🛰️

Elon Musk's xAI released Grok 4.7 on Monday, calling it "a notable improvement over Grok 4.6 at the same price and speed," after Musk pushed the launch timeline back at least five times since late July — from "four weeks out," to "a few weeks," to "3 to 4 weeks," to "10 days" on September 1, and finally "needs a few more days to cook" on September 11. Musk posted on X that Grok 4.7 is "a strong combination of intelligence, speed & low cost," and the model is live immediately in the Grok app, Cursor, Grok Build, and the xAI API with no waitlist.

Grok 4.7 carries 2.1 trillion parameters, up 40% from the 1.5 trillion in Grok 4.6, which itself was a refinement of Grok 4.5. xAI set pricing at $2 per million input tokens and $6 per million output tokens. The company also folded in supplemental training data pulled from SpaceX — Starlink satellite telemetry, manufacturing records, and engineering failure logs — framing the model as better suited to reasoning about hardware and physical systems than text-only models.

Benchmark scores placed Grok 4.7 behind the leading competitors across multiple tests. On GDPval, which scores economically valuable knowledge work on an Elo rating, Grok 4.7 posted 1695 against Claude Fable 5.1's chart-topping 1735. On AA-Briefcase, which measures multi-hour office workflows, Grok 4.7 reached 1657 versus Fable 5.1's 1678. On CursorBench 4.0, Grok 4.7 landed mid-pack — more expensive per task than GPT-6 Astra and Claude Sonnet 5, but still trailing Fable 5.1 at every price point.

The release follows a familiar pattern for xAI: Grok 4.5 launched in July with the industry's largest training cluster and third-place scores behind Claude and OpenAI, Grok 4.20 prioritized speed and personality over reliability, and Grok 4.6 trailed the frontier on coding autonomy. Grok 4.7 lands in the same bracket, positioned as xAI's most capable coding and knowledge-work model but not at the top of the leaderboards.

The launch comes just over a month after Grok 4.6 and ahead of upcoming model releases from OpenAI and Anthropic, with SpaceXAI's announcement describing Grok 4.7 as twice as fast and half the price of its predecessor while incorporating a longer reinforcement learning run on harder, multi-hour tasks.