Cursor Router auto-picks your model, cuts coding cost 30–50%
On by default for Teams, it routes each prompt to Cost, Balance, or Intelligence tiers; admins get allow/block lists and A/B tests showed up to 60% savings.
Copy markdown
Auto mode now picks the model for you
Cursor Router is a request-level classifier that reads each prompt's complexity and sends it to the best-fit model — and it's on by default for Teams plans. If you build on a team, your requests are already being routed unless an admin turns it off.
30–50% cheaper, up to 60% in A/B tests
Cursor's own numbers: 30–50% lower cost versus Opus 4.8 routing on high-volume accounts, and 60% savings across millions of A/B requests. Cost per commit came in at $6.76 in Intelligence mode and $4.63 in Balance — and you're billed at the routed model's standard rate.
Three dials: Cost, Balance, Intelligence
Pick your quality-vs-spend tradeoff. Intelligence targets frontier quality, Cost maximizes token efficiency, and Balance is the daily-driver middle. The classifier, not you, decides which underlying model each request actually hits.
Admins get allow/block lists — set them
Team admins can restrict modes, set a default, and allow or block specific underlying models per team or group. The catch: leave it untouched and you hand per-request model choice to the classifier, so pin an explicit policy if your workflow depends on one model.