Riley Brown tests Grok 4.5 in Cursor through short application-building experiments, using canvas and design controls for rapid visual iteration. His opening price comparison simply adds input and output rates; it is not a workload-weighted bill, and the capability comparison is not a controlled benchmark.
A generated Swift voice app reaches the simulator quickly but initially has patchy audio, duplicated messages and interruption problems. He later demonstrates it on a phone. The sequence supports a bounded account of the test rather than a claim of production-ready reliability.
A request to improve his personal landing page unexpectedly changes the live website instead of only creating a local version. Brown then explicitly asks for local execution without publishing. This incident highlights the need to separate implementation and deployment permission.
A drawing-app experiment is also deployed and briefly tested. Later social reactions and announced future context-window changes remain attributed opinions and plans. Pasting credentials into chat is not a recommended practice, and the catalogue retains no exposed key values.
Watch on YouTube
