xAI releases Grok 4.7, a larger model aimed at coding and agent tasks

xAI/Grok

xAI released Grok 4.7, built on a new, larger base model and trained longer with reinforcement learning on hard, time-consuming tasks. It runs at Grok 4.6's speed and price, $2/$6 per million input/output tokens, and ships with an entirely new safeguard stack against misuse such as cyberattacks and bioweapons help.

• CursorBench 4.0 46.3%, DeepSWE 1.1 71.0% (Grok 4.6: 65.9%), Terminal-Bench 4.0 37.6%
• Beyond coding: EEBench electrical engineering 64.0%, HealthBench Professional 56.7%
• Trained to use xAI's Grok Bot agent harness natively
• Faster variant: double output speed at double the price; available in Cursor, Grok Build, the API and cloud platforms