Qwen3.8 Flash vs DeepSeek 4.1 Flash draw a coffee Venn

In short

Qwen gets all three runs right and DeepSeek only one.

TikZ: a three-set coffee Venn diagram, by deepseek-v4.1-flash
DeepSeek 4.1 Flash, run 1. Every drink in its region, and six of the eight labels across an edge.$0.02SVGPNGPDFvenn.tex
TikZ: a three-set coffee Venn diagram, by qwen3.8-flash
Qwen3.8 Flash, run 3. One render, and every label clear of every line.$0.04SVGPNGPDFvenn.tex

Finding

Both models knew where every drink goes, in all six runs. Qwen3.8 Flash kept every label clear of the circle edges in three of three runs, and DeepSeek 4.1 Flash did in one of three. The difference in how they worked is that Qwen measured the layout in Python, and DeepSeek never did.

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.

Checks

checks
check · how it is decidedDeepSeek 4.1 Flashrun 1DeepSeek 4.1 Flashrun 2DeepSeek 4.1 Flashrun 3Qwen3.8 Flashrun 1Qwen3.8 Flashrun 2Qwen3.8 Flashrun 3clock stop
Compilesmeasured · pdflatex builds the file offline.✓ pass✓ pass✓ pass✓ pass✓ pass✓ pass
Seven regions, three namesmeasured · Three circles found in the PDF with every overlap present, and each set name placed by its circle.✓ pass✓ pass✓ pass✓ pass✓ pass✓ pass
Every drink in its regionmeasured · Each of the eight words found in the PDF, and the centre of each one inside the region it belongs to.✓ pass✓ pass✓ pass✓ pass✓ pass✓ pass
No label on an edgemeasured · No letter touches a circle's outline, with the line's width counted.✗ fail✗ fail✓ pass✓ pass✓ pass✓ pass

A run marked as stopped was cut off before it finished, and is scored on its last rendered frame. Ground truth: 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. Every model on this task.

The sorting is the easy half. An iced latte has espresso, milk and ice, so it goes in the middle, and a pour-over has none of them, so it goes outside. Both models got all eight drinks right every time. The hard half is fitting the words in, because the overlaps are small lens shapes and "iced long black" is wide. DeepSeek put labels across the circle lines in two of its three runs, and Qwen worked out where each label would land in Python first and got it right every time.

Verdict

Verdict

Qwen3.8 Flash passes this in three of three runs and DeepSeek 4.1 Flash in one of three. The next thing to try is four sets, which three circles cannot draw.

n = 3 runs a model. p = 0.4, Fisher exact, after the runs.

The six runs cost $0.18: $0.11 for Qwen and $0.07 for DeepSeek.

Older, 24 Sept

DeepSeek found the grader and ran it

Newer, 25 Sept

Qwen3.8 Max Prime draws three coffee shrubs