Cursor Adds Grok 4.7 for Long-Running Coding Tasks
Cursor announced on September 21, 2026 that Grok 4.7, xAI’s newest model, is now available inside Cursor. xAI describes Grok 4.7 as its most capable model yet for long-running coding and knowledge work, and Cursor is one of the first coding tools to offer it alongside the Grok API, Grok Build, and other third-party platforms.
Details
- Built for extended tasks: xAI says Grok 4.7 uses an expanded base architecture and additional reinforcement learning compared to Grok 4.6, aimed at self-verification over long, multi-step sessions
- Coding benchmarks: xAI reports 46.3% on CursorBench 4.0, 71.0% on DeepSWE v1.1, and 37.6% on Terminal-Bench 4.0
- Pricing: $2 per million input tokens and $6 per million output tokens, with a faster variant available at double the price for double the output speed
- Safety note: xAI cites HackerBench v0.3 results showing the model complies with only 3.3% of risky prompts
- Availability: ships in Cursor alongside existing model options, with no separate waitlist mentioned
What happened next
Grok 4.7’s launch triggered near-simultaneous rollout across coding tools; GitHub Copilot added the model the same week, and Cursor’s addition means developers on either platform can pick it for agentic coding work without waiting on a broader release. Cursor gave no indication of retiring Grok 4.6 access, so both versions currently sit side by side in its model picker.