Grok 4.5 hits Cursor with $2/$6 pricing

Teams that live in the IDE now have another frontier model to compare against Claude and GPT. Cursor says xAI’s Grok 4.5 is already live in desktop, web, iOS, CLI, and SDK, with public release set for July 9.

Cursor shipped Grok 4.5 in the IDE with $2/$6 pricing and published benchmark scores.

What changed

Cursor and xAI described Grok 4.5 as a mixture-of-experts model trained on trillions of Cursor interaction tokens, not just static code data. The pitch is broader than coding: the model is positioned for STEM, research, and knowledge work, while Cursor’s own coding specialist, Composer 2.5, stays available.

The launch also came with specific performance and pricing details. Cursor published scores including 83.3% on Terminal-Bench 2.1 and 64.7% on SWE-Bench Pro, while listing base API pricing at $2 per million input tokens and $6 per million output tokens. A faster tier is priced at $4 and $18.

• Available in Cursor desktop, web, iOS, CLI, and SDK
• Public xAI release follows on July 9, 2026
• Cursor says Grok 4.5 is the first jointly trained SpaceXAI + Cursor model
• Cursor also warned that an earlier codebase snapshot may have inflated CursorBench results

Why it matters

For developers, the practical question is not just benchmark rank. It is whether Grok 4.5 can cut task cost and latency inside real coding workflows, especially for long agent runs, repo edits, and tool use.

The pricing matters because it puts pressure on rivals to justify higher token costs with better accuracy or fewer retries. If teams can get near-Opus performance at lower spend, procurement decisions may shift from model prestige to per-task economics.

The benchmark caveat matters too. Cursor noted possible contamination in CursorBench from an earlier repo snapshot, so the strongest claims still need outside replication before teams treat them as buying signals.

The bigger takeaway is simple: Grok 4.5 is no longer a beta rumor. It is now a priced product inside a major IDE, and the next fight is whether independent tests match the launch numbers.