Not one model. In my 5-test Cursor run, Gemini won 4 tests (on low effort; medium/high errored), Claude built the best landing page, and Grok high was slowest and not worth the wait.

Video: https://www.youtube.com/watch?v=I4qA5bXzHpM (@janekzarin)

Summary: Cursor recommends Grok by default. I wanted to check that on normal, day-to-day work, the kind that actually helps a small business sell: a shop calculator, a landing page, an edit that doesn't break the page. Not an official benchmark, just my first-impression ratings.

What I tested

Five models side by side in Cursor: Grok 4.7, Claude Opus 5, Gemini, Muse and GPT. Each got a completely fresh folder, so it could only use what it generated itself. Same one-shot prompts:

  1. A single-file HTML game
  2. A business calculator for an online shop (VAT, taxes, costs)
  3. Drawing with code: an SVG of a moose riding a scooter through a forest
  4. A single-file landing page for a cordless garden tool shop
  5. Edit without breaking: add a currency switcher with a fixed rate to that shop

It is not the fairest setup. Effort levels differed: Grok and Muse on high, the others started on medium.

What broke

  • Gemini kept erroring in Cursor on medium and high, two or three times in a row. Low effort worked, so most Gemini results are on low.
  • My own mistake: on the first calculator run I forgot to switch the model, so the "Gemini" folder actually had Grok code. I made a new folder and re-ran it with real Gemini.
  • Grok high was slow. On the game it finished last and the result looked about like Claude's. Grok high-fast was much faster on the next test.

Results (my ranking)

  • HTML game: Gemini first, then Claude. Grok's was similar to Claude's but not worth the wait.
  • Shop calculator: Gemini (low) was the cleanest, one I'd use and not be afraid to sell. ChatGPT looked the most "sellable as a product"; Grok and Muse did nice simple ones.
  • SVG drawing: Gemini 1, Claude 2, ChatGPT 3, Muse 4. Grok lost this one.
  • Landing page: Claude best (working links, Q&A section, animation), Muse second, Gemini third, Grok fourth (more blog than shop), ChatGPT last (colors didn't match).
  • Currency switcher edit: Gemini first, Claude second, Grok third with its own unique solution. ChatGPT's switcher worked but was really ugly.

The marketing takeaway

If you use AI to build things that get you sales (calculators, landing pages, quick page edits), don't just take the default model. For a one-shot shop page, Claude gave me the most ready-to-use result. For quick tools and edits, Gemini did best even on low effort. Pick the model per task, and do a second prompt before you judge a page.

At the end of the video I show this month's Cursor usage. Other models ran four against one Grok, so the percentages aren't one-to-one.