ଗ୍ରୋକ୍ 4.7

Grok 4.7 is SpaceXAI's flagship for coding, agentic tasks and knowledge work. It accepts text and image inputs in a 500,000-token context, supports function calling and structured outputs, and provides reasoning efforts from low through xhigh, with high as the direct API default.

xAI built Grok 4.7 around a larger base model and longer reinforcement learning on difficult, extended tasks. The lab says it checks its work more carefully and handles long context better, with particular attention to coding and professional work.

The launch comparison shows progress without claiming a clean sweep. xAI reports 46.3% on CursorBench 4.0 and 37.6% on Terminal-Bench 4.0; its table still places Fable 5.1 ahead on both. That makes the relevant question whether the model finishes your particular workflow reliably, rather than whether it leads every published test.

xAI also describes a redesigned safeguard system intended to resist jailbreaks while preserving legitimate defensive security work. These are the lab's reported results. Evaluate useful task completion and the handling of sensitive requests together when deciding where the model fits.