6 questions · about 3 minutes
Which model wrote this?
Six things that actually happened when two models got the same brief. Guess which one did it — the answer is on the test page it came from.
Question 1
Two models were asked for a blog post page, with a sample post in the folder. One of them produced 46 KB. The other produced 4.4 MB, because it generated its own hero, author and thumbnail photographs. Which one went and made the pictures?
Gemini. 4.3 MB of the 4.4 was images it drew itself — nobody asked for any. GPT spent its seven minutes writing three further article pages instead, so that the internal links the brief asked for had somewhere to go.
Question 2
An idle factory game, one HTML file, seven production stages. One entry came back at 582 lines and the other at 2,530 — four times the length, almost all of it in the drawing. Which model wrote the long one?
Claude, at Max effort. It looks considerably better and it is the one that slows down late in a run — while GPT’s plainer 582 lines never speeds up either, because no upgrade makes its machines faster. Both built every stage in the brief.
Question 3
Two versions of Gemini Flash got the same SaaS landing page brief: single file, no external assets. One stayed self-contained. The other fetched two typefaces from Google. Which one went shopping?
The older one. 3.6 also took considerably longer and said it had tested the result — and its page scrolls 33 px sideways on a 390 px phone. 3.7 answered fast, tested nothing, and fetched nothing.
Question 4
“Invent an original mechanic,” both were told. One opened with a single line — “One button. Two orbits. No brakes.” The other opened with a three-panel tutorial: GLIDE, TETHER, SLING. The brief said the mechanic had to be understandable within five seconds. Who wrote the tutorial?
Gemini — 1,868 lines against ChatGPT’s 264, and a mechanic that needed a manual. It is the more ambitious game of the two and the harder one to hold: its tether grabs the anchor nearest the ship rather than the one nearest your cursor.
Question 5
A $10,000 agency homepage, single file. One entry’s portfolio is three photographs pulled from Unsplash, which are empty frames the moment the network goes. Which model reached for the stock photos?
The Pro, in 781 lines of enormous typography and not one button. The Flash drew 24 pieces of inline SVG instead and fetched no images at all — and then attached a costing tool and four world clocks to the page.
Question 6
Two Gemini versions, the same brief, minutes apart. One opened its page with “Your work moves at the speed of thought. FlowPilot runs the rest.” What did the other one open with?
Almost the same sentence, one verb apart. Two different versions of the model, given complete creative freedom over a headline, wrote the same twelve words — which is the most quietly interesting thing in that round.
Your result
Answer all six and your score appears here. Each one tells you the answer as you go.