SpaceXAI released Grok 4.7 on 21 September, its new flagship for coding and what the company calls knowledge work. It is served at the same prices as Grok 4.6, at $2 per million input tokens and $6 per million output, with a faster variant at double the price. It is available in Cursor, Grok Build, the Grok API and third-party coding harnesses.
The published benchmark table is unusually candid about where the model does not lead. Grok 4.7 scores 46.3% on CursorBench 4.0 against 51.8% for Fable 5.1, 38.0% on Terminal-Bench 4.0 against 57.9%, and 56.7% on HealthBench Professional against 62.1%. It does lead the four models compared on EEBench, at 64.0%, and by a wide margin on the Harvey Legal Agent Benchmark, at 19.6% against 6.7% for Fable 5.1 and 2.5% for GPT-5.6 Sol.
So the headline claim of half the price is about rivals’ list prices, GPT-5.6 Sol at $4 and $20 and Fable 5.1 at $10 and $50 per million tokens, rather than a cut to SpaceXAI’s own. On cost per task that is a real argument. On raw capability, the company’s own figures put Grok 4.7 behind the frontier on most of the coding and clinical tests it chose to publish.
SpaceXAI has also begun giving selected cybersecurity partners access to the model’s red-team capabilities.