Summary
On July 22, 2026, Cursor introduced three optimization modes for its Auto mode, which is powered by the Cursor Router: Intelligence for frontier quality, Balance for strong everyday quality, and Cost for optimized token spend. The router classifies each request by task type and complexity to pick a model, and admins get controls to standardize teams on Auto mode and govern model access.
What changed
Cursor shipped Intelligence, Balance, and Cost optimization modes for its router-driven Auto mode, with per-request task/complexity classification and admin controls for defaults and model access.
Why it matters
As multi-model coding tools proliferate, routing and cost governance become the real product surface, not the underlying model. Giving teams an explicit quality-versus-cost dial plus admin standardization directly targets enterprise spend predictability, a top blocker to scaling coding agents, and positions Cursor against GitHub Copilot's auto model selection and Windsurf/Devin's routing.
Evidence excerpt
Cursor introduced new optimization modes: Intelligence (frontier quality), Balance (strong quality for daily use), and Cost (optimized token spend), along with per-request classification by task type and complexity, and admin controls for standardizing on Auto mode.