I Tested Every Cursor Model on Normal Work · published 1 November 2026
Play at 0:13 Watch on YouTube Cursor
How did I compare the Cursor models for ordinary work?
I ran the same one-shot prompts side by side and I ranked them by my first impression. The effort levels were not matched, so this is not the official benchmark.
The picker that day was recommending Grok, and Grok 4.7 was the default I saw. I still wanted a model per kind of job. The five I ran were Grok 4.7, Claude Opus 5, Gemini, Muse, and GPT. Grok started on high. Muse was on high. The others were on medium. Each model got a new folder, so it could not reuse files already on the computer. The five tests were a single-file HTML game, a shop calculator with tax, a drawing made of code, a landing page, and an edit that had to leave the page working. I started them at different times, so I did not record a fair speed winner. With five days left in the month I was already at 48%, mostly from Grok Bot, and I expected that to push me onto Ultra. I ran the test anyway.