Grok 4.7 (xhigh)

Released by SpaceXAI on September 21, 2026, Grok 4.7 is a frontier language model optimized for agentic coding, professional knowledge work, and multi-hour reasoning tasks. Building on the architecture of its predecessor, Grok 4.6, the model utilizes a significantly larger base and underwent an extended reinforcement learning phase specifically designed to improve self-verification and long-horizon consistency. The model is released in two primary variants: Grok 4.7 (high) and the higher-intelligence Grok 4.7 (xhigh), both of which are integrated into the Grok Build and Cursor ecosystems.

Grok 4.7 introduces a 500,000-token context window, allowing it to manage expansive codebases and complex documentation sets with high reliability. In technical benchmarks, the model demonstrated substantial gains in autonomous engineering, scoring 46.3% on CursorBench 4.0 and 71.0% on DeepSWE v1.1. Its performance on Terminal-Bench 4.0, which measures a model's ability to operate in live terminal environments over extended durations, nearly doubled compared to earlier iterations, reaching 38.0%.

Architecturally, the model is estimated to feature approximately 2.1 trillion parameters. Beyond its coding capabilities, SpaceXAI trained Grok 4.7 to natively understand the Grok Bot harness, enhancing its performance in conversational reasoning and general-purpose knowledge retrieval. The model also incorporates a redesigned safeguard stack aimed at providing high resistance to jailbreaks and cybersecurity risks while maintaining low refusal rates for legitimate research and development queries.

For optimal results in agentic workflows, the model is designed to persist through difficult, long-running assignments without premature abandonment. Users of the Grok Bot harness can leverage its native understanding of the system to perform complex, multi-step professional tasks such as document synthesis, legal analysis, and sophisticated software architecture design.