Cursor Router Uses Compass Complexity Scoring

Compass points toward a future where model routing optimizes for outcomes, not just prompt length or task labels.

• –Satisfaction prediction gives routing a feedback loop tied to user success
• –Complexity scoring can reserve expensive frontier models for genuinely difficult coding work
• –The reported 96% signal rate is promising, but comes from an individual post rather than a published evaluation
• –Developers may gain lower costs without manually switching models, though opaque routing makes debugging harder
• –Cursor’s broader router supports Cost, Balance, and Intelligence modes across its clients