The source says Grok 4.7 is available today in Cursor, Grok Build, the Grok API, third-party coding harnesses, model routers, and cloud platforms. It lists pricing from $2 per million input tokens and $6 per million output tokens, with a faster variant at twice the output speed and twice the price. It also reports 46.3% on CursorBench 4.0, 71.0% on DeepSWE v1.1 at high effort, and 37.6% on Terminal-Bench 4.0. These facts describe the vendor’s release and reported evaluations, rather than independently established production performance.
The mechanism is a larger base model trained through a longer reinforcement-learning run on harder, multi-hour tasks, paired with training for self-verification, longer-context management, and native understanding of the Grok Bot harness. Why now is that deployment is not deferred: the same announcement pairs the capability claim with access in Cursor and common API and harness distribution channels. The immediate implication is that teams selecting models for agentic coding can test a potentially lower-cost long-horizon option without waiting for a separate platform rollout. A second implication is that model choice should be measured in completed-task cost and reliability, not token price or a single benchmark alone.
The evidence boundary is that all performance, safety, and price-performance comparisons in this record come from SpaceXAI’s own announcement and can support a release record, not a conclusion of frontier leadership. The first watch condition is whether independent long-horizon coding evaluations and production users reproduce the claimed reliability and cost advantage. Also watch whether availability and quoted pricing remain consistent across the API, Cursor, routers, and cloud providers, and whether the stated safety behavior is corroborated by external testing.