CursorBench

This page may contain stale information. Last updated: 2026-09-22

Definition

CursorBench is cursor’s benchmark suite for evaluating coding model performance on longer-running, real-world tasks inside the IDE. CursorBench 4.0 stresses multi-step coding workflows and plots accuracy against cost and token count.

Key Points

• 2026-09-21: grok 4.7 claims frontier price-performance on CursorBench 4.0 at $2/$6 per M tokens (2026-09-21-grok-4-7-release)

CursorBench scores are vendor-reported by Cursor/SpaceXAI. Independent replication pending.

• Decrypt coverage: Grok 4.7 mid-pack on price-performance; Claude Fable 5.1 leads at every price point on the chart

Related

• coding-agents
• ai-benchmarks
• frontier-models
• grok
• cursor
• spacexai

Sources

• 2026-09-21-grok-4-7-release
• 2026-09-21-grok-4-7-decrypt
• 2026-09-21-grok-4-7-cursor-blog