Finding
Some models have the spoon flat on the surface to stir and others have the spoon leaving the cup while it stirs.
I asked three models for a cup of coffee with a spoon stirring in it, in HTML and CSS. This is as I want more dynamic behaviour checks on the site.
Sonnet stirs it in all three runs.
GLM stirs, but the spoon leaves the cup.
DeepSeek's spoon is lying on the surface.
| model | runs | HTML and CSS only | A CSS loop that moves something | Reads as a cup seen into | The spoon does the stirring | Reads as inside the cup | Handle stays on the bowl |
|---|---|---|---|---|---|---|---|
| Claude Sonnet 5.5 | 3 | 3 of 3 | 3 of 3 | 3 of 3 | 3 of 3 | 3 of 3 | 3 of 3 |
| GLM 5.3 FlashX | 3 | 3 of 3 | 3 of 3 | 3 of 3 | 3 of 3 | 0 of 3 | 3 of 3 |
| DeepSeek 4.1 Flash | 3 | 3 of 3 | 3 of 3 | 3 of 3 | 0 of 3 | 2 of 3 | 1 of 2 · 1 n/d |
Each cell is runs that passed that check, of runs where it was decided. Every run, with pictures.
| Failed | How many | Dimension | Property | Mode |
|---|---|---|---|---|
| The spoon does the stirring | DeepSeek, 3 runs | function | behaviour | missing |
| Reads as inside the cup | GLM, 3 runs | validity | relation | transposed |
| Reads as inside the cup | DeepSeek run 1 | validity | relation | transposed |
| Handle stays on the bowl | DeepSeek run 1 | validity | geometry | distorted |
Verdict
DeepSeek's failure is a missing behaviour, the failure of GLM is that the spoon leaves the cup while stirring.
n = 3 runs a model. No test.
The nine runs cost $0.96.