Gallery, prompt v1
A three-set coffee Venn diagram
One frozen prompt and every model that has run it, newest first.
The prompt, v1, frozen
Using LaTeX and TikZ, draw a Venn diagram of three overlapping circles labelled "Has espresso", "Has milk" and "Served cold". Write each of these drinks in the region where it belongs: short black, flat white, iced latte, iced long black, cold brew, babyccino, milkshake, pour-over.
Every model, its first three runs.
One square a check, in this order:
- Compiles
- Seven regions, three names
- Every drink in its region
- No label on an edge
passed, failed, undecided.
| Claude Sonnet 5.5 | |||
| Opus 5.5 | |||
| GPT-6 Sol | |||
| Perceptron Mk1.5 | |||
| Command A+ | |||
| Kimi K3 | |||
| Ember 1 | |||
| GPT-6 Luna | |||
| GLM 5.3 FlashX | |||
| Grok 4.7 | |||
| Gemini 3.8 Flash | |||
| MiMo v2.6 Pro | |||
| Qwen3.8 Flash | |||
| DeepSeek 4.1 Flash |
How it is checked
Short black is espresso only, flat white espresso and milk, iced long black espresso and cold, iced latte all three, babyccino milk only, milkshake milk and cold, cold brew cold only, and pour-over outside every circle. The circles and words are read back out of the compiled PDF, each word has to be centred in its own region, and no letter may touch a circle's outline, with the line's width counted.
When this saturates: Four sets, which three circles cannot draw, so the model has to know to use ellipses.
Failure types seen here: labelling (labels that don't match what is drawn). The same types across every task.
The rules behind all of thisThis task's prompt, checks and any checker, on GitHub
run 1 of 25 Sept, 4 rounds, $0.06Every drink right and every label clear, with flat white close to both lines. It said the render looked correct.
run 2 of 25 Sept, 3 rounds, $0.07Every drink right, and iced latte across a line.
run 3 of 25 Sept, 3 rounds, $0.06White boxes behind every label, and every letter clear of the lines.
run 1, 7 rounds, $0.04Every drink in its region and clear of every edge. Round 5 pushed iced long black onto a line and round 6 moved it back.
run 2, 3 rounds, $0.03Passed on all three of its renders, with a geometry script checking the label positions along the way.
run 3, 1 round, clock stop, $0.04One render, and it passes all four. It measured each label in LaTeX and its clearance from each circle in Python first, and the clock ended it while it was still tuning the layout.
run 1, 2 rounds, $0.02Every drink in the right region, and six of the eight labels lying across a circle's edge.
run 2, 3 rounds, $0.03Two labels across an edge, with white boxes behind them that hide the line they cross.
run 3, 5 rounds, $0.02Every drink in its region and clear of every edge, in five rounds.







































