Today, we are releasing Grok 4.7. It’s our most capable model yet for coding and knowledge work, and it’s available today in Cursor!
Grok 4.7 works longer on difficult tasks, checks its own work more carefully, and ships with the best-calibrated safeguards to date. It’s served at the same price and speed as Grok 4.6.
Highlights of Grok 4.7:
-
Grok 4.7 scores 46.3% on CursorBench 4.0, up from 40.4% for Grok 4.6, and nearly doubles Grok 4.6 on Terminal-Bench 4.0 (20.3% → 38.0%). It improves on every benchmark, including AA Briefcase, Harvey Legal Agent Benchmark, HealthBench Professional, and EEBench.
-
It was trained with a longer reinforcement learning run on a harder mix of tasks, weighted toward problems that take many hours to complete. It’s better at verifying its own work and managing longer context.
-
Grok 4.7 was trained to natively understand the Grok Bot harness, making it better at conversational tasks and general knowledge work, not just code. It’s stronger at creating documents and presentations.
-
Grok 4.7 was built with an entirely new safeguard stack. It’s the strongest Grok model tested on refusals and jailbreak resistance, while keeping refusal rates low for legitimate security and research work.
Grok 4.7 is available today in Cursor and Grok Build]. It is also available through the Grok API, third-party coding harnesses, and model routers and cloud platforms. Grok 4.6 remains available, and Composer is still our fast, cost-efficient model for everyday coding.
Read more in the announcement. We’d love for you to share your first impressions.