elvis

Sasha Rush

TestingCatalog News 🗞

Cursor has launched Cursor Router, a model routing system that analyzes each incoming coding request and dispatches it to the most appropriate model. Complex tasks go to frontier models; simpler edits go to cheaper ones like Grok 4.5. The result, according to Cursor, is frontier-quality output at 60% lower cost.
The router is available across desktop, web, iOS, CLI, and SDK. Developers can choose between three modes - Intelligence, Balance, and Cost - depending on how much they care about output quality versus token spend. Enterprise admins get additional controls: they can enforce Auto mode across teams, restrict specific models, and set group-level token budgets.
The broader point here is interesting. Routing is increasingly where the real leverage is in AI infrastructure. A well-tuned router doesn’t just save money - it can potentially outperform any single model by matching task type to model strengths. That’s a different kind of capability gain than just waiting for the next frontier release.
One open question: should routing logic be something you hand off to a vendor, or something you control yourself? Different teams have different cost/quality trade-offs, different model access, different latency requirements. An open-source router you can tune yourself would be genuinely useful here - though nothing obvious has emerged yet to fill that gap.

Collections

Cursor has launched Cursor Router, an automatic model routing system that selects the most appropriate AI model based on the task being performed. The system claims to deliver results on par with frontier models while reducing costs by 60%.

Cursor Router automatically picks the right model for each task

Let c… when we used cursor auto mode long back its burning tokens like hell