xAI Releases Grok 4.7 LLM for Advanced Coding and Knowledge Tasks
xAI has announced Grok 4.7, a flagship large language model built on a larger base architecture with extended reinforcement learning for multi-hour coding and knowledge work. The model starts at $2 per million input tokens and $6 per million output tokens, promising twice the speed of comparable models at half the price. Grok 4.7 delivers strong benchmark performance in agentic software engineering and specialized knowledge tasks while significantly reducing API costs for long-running workflows. Its native training on agent harnesses and direct integration into tools like Cursor accelerate the practical use of autonomous AI software developers. The model features improved self-verification and extended context management, blocking 96.7% of high-risk dual-use prompts on the HackerBench v0.3 cybersecurity test. It is accessible via the Grok API, Cursor, and Grok Build, with an optional high-speed tier offering double the output generation speed at twice the cost.
## BACKGROUND
Reinforcement learning (RL) combined with longer reasoning loops allows modern AI models to self-verify code and persist across complex multi-step tasks. Benchmarks like CursorBench evaluate AI agents on real-world multi-file programming tasks executed within developer workflows to measure practical software engineering capabilities.