Cursor Publishes Router Internals: a Complexity Predictor Called Compass, and 68% Cost Reduction Claims
Cursor's August 6 engineering post details a two-stage router: 'Compass' assigns each turn a 0–1 complexity score by predicting whether the user will be satisfied, then a taxonomy across domains (backend, frontend, database), tasks (bug fixes, commands, tests), and modifiers (visual changes, product questions) picks the model. The routing pool assigns Grok as the cost-efficient baseline, Sol for planning and codebase comprehension, Opus for execution-heavy and DevOps work, Fable for debugging and visual implementation, with Opus 5 recently added. Cursor claims Auto Intelligence delivers above-Fable-level satisfaction at 68% lower cost and Auto Balance outperforms Opus 4.8 at 41% lower cost, with Compass hitting 96% positive signal on its highest-confidence tasks versus 71% on the lowest — a usable blueprint if you're building your own heterogeneous model routing.
Source
↳ Follow the thread