SpaceXAI: Grok 4.7

Grok 4.7 is xAI (SpaceXAI) frontier model released on September 21, 2026, positioned as the successor to Grok 4.6 for coding, agentic workflows, and professional knowledge work. It is served as grok-4.7 through the Grok API, Grok Build, Cursor, and third-party gateways, with a 500,000-token context window, text and image input, function calling, web and X search, code execution, and configurable reasoning effort from low through xhigh. Pricing starts at $2 per million input tokens and $6 per million output tokens for prompts under 200K tokens, with higher tiers for longer context.

The model builds on a larger base than Grok 4.6 and was trained with a longer reinforcement-learning run on harder, multi-hour tasks, with emphasis on self-verification and long-context management. Supplemental training on anonymized Cursor workflow data improves IDE-style and harness-native agent behavior. Official evaluations highlight strong agentic software engineering (including DeepSWE v1.1 at 71.0% pass@1 at high effort), competitive GDPval-AA and office-work ratings, and improved cyber capability probes such as CyberGym (80.3% mean reproduced at high effort) alongside a redesigned safeguard stack.

Grok 4.7 is proprietary and not released as open weights; deployment is API- and product-integrated only. It targets engineers, researchers, and knowledge workers who need a single model for long-horizon coding, terminal agents, document and presentation tasks, and multimodal understanding, with reasoning modes that trade latency and cost for depth on difficult problems.