How each model did
Ranked by our pick, then checks passed, then cost. Time and cost count every try; each dot is one check, in the order listed below (filled: passed, half: partly, empty: failed). Click a model for its notes.
Notes on each answer
1
Muse Wonder Full
Our pick
3.5 of 4
Made in Muse Code, not over the API: Full could screenshot its own work and fix it, and took 3 rounds. The only tulip: the espresso settles, the milk builds a base, five petals are pushed in one after another, and a pull-through cuts the stack, leaving a small heart on top. It is small in the cup and still close to a leaf if you are strict, and the crema is dotted with large spots, so the cup looks less real than Opus's. The light glints across the surface. Its fixes made the hearts rounder and wider apart so that it read as a tulip.
- ◐ The crema and the cup look real (partly)
–
The most realistic cup: a white cup and saucer on dark wood, a crema with real texture, and a steel jug pouring from below. But the milk lands as a lumpy white column with jagged edges, not a tulip, and it sits flat on the crema. The light moves across the cup rim and the crema. 6.6 minutes, $0.78.
- ✗ The finished pattern is a tulip: stacked petals with a pull-through
6.6 min · 39,176 output tokens (26,407 thinking) · $0.784
Titled "Tulip", with steam rising and the jug pouring from above, but it pours a large leaf, a rosetta with a centre line and nested chevrons. Clean and pretty, and the wrong pattern. Our first two requests were refused before it wrote anything (by our account's credit limit at its listed 943,718 tokens of room, then over a thinking setting this endpoint does not take); the third, with 128,000 tokens, is the answer. 1.1 minutes, $0.18.
- ✗ The finished pattern is a tulip: stacked petals with a pull-through
- ◐ The crema and the cup look real (partly)
- ◐ The light follows the pointer across the surface (partly)
67 s · 8,868 output tokens (3,274 thinking) · $0.178 for this answer; 67 s and $0.178 over 3 tries
Took 3 tries: failed at 943,718 tokens; failed at 943,718 tokens, low thinking; finished at 128,000 tokens.
One answer per model, at each model's default settings unless the card says otherwise, through OpenRouter. If an answer ran out of room or stopped short, we asked again with more room, then with thinking off, and say so. This is a quick look, not a benchmark: run it again and every answer will differ. For measured results, see our benchmarks.
How we pick. Claude Opus 5.5 tried every answer (clicking, dragging and playing it in a browser), marked the checks and drafted the notes. Our pick is a judgement among the answers with the most checks passed, not a score: the checks show what works, the pick is which of those we would rather use. Opus is an Anthropic model, and Claude answers are our pick in 17 of 28 tests, so weigh Claude picks with that in mind. GPT-6.1 Sol, shown anonymised stills and our check marks, put Muse Wonder Full first as the only recognisable tulip, then Claude Opus 5.5 for its cup and crema, then Muse Wonder, whose pattern it read as a leaf. We agree.
What we checked, by trying each answer ourselves: Shows the pour from the jug, with the milk landing in the espresso; The finished pattern is a tulip: stacked petals with a pull-through; The crema and the cup look real; The light follows the pointer across the surface. A partial pass counts as half.
Verdict
Muse Wonder Full is our pick because it is the only answer that pours a tulip: stacked petals with a pull-through, made in three rounds in Muse Code, where it looked at its own page and reshaped the hearts. Claude Opus 5.5 has by far the most realistic cup and crema, but its milk lands as a lumpy blob, and Muse Wonder poured a leaf. Full's advantage came from being able to look and fix, which the other two, with one try each over the API, could not.
