Gallery, prompt v2
Exploded French press
One frozen prompt and every model that has run it, newest first.
The prompt, v2, frozen
Draw an exploded assembly diagram of a French press, as an SVG. Use the standard three-piece plunger filter rather than a two-piece travel press.
Every model, its first three runs.
One square a check, in this order:
- All seven parts
- Stack order
- Filter fits the glass
- One assembly axis
- Labels match
passed, failed, undecided.
| Pareto 26.10 Preview | |||
| Opus 5.5 | |||
| Pareto | |||
| Claude Sonnet 5.5 | |||
| Perceptron Mk1.5 | |||
| Command A+ | |||
| Sol | |||
| GLM 5.3 FlashX | |||
| DeepSeek 4.1 Flash | |||
| Qwen3.8 27B |
How it is checked
The standard three-piece plunger, from the top down: spiral (spring) plate, mesh screen, cross-plate. The cross-plate is the bottom piece, because its threaded stem is what the rod screws into. The spiral plate above the mesh is what presses the mesh edge out against the glass. Checked against the WebstaurantStore anatomy diagram, which is drawn exploded and labelled, and against each part's y coordinate in the SVG rather than its comment. Two-piece travel presses exist and stack differently, which is why v2 of the prompt names the design.
A check decided by eye carries the rule I apply to it. I wrote those rules down on 21 September 2026, after the first results were scored, so for those results they record how I made the calls after the fact.
When this saturates: Draw it in section rather than exploded, so the mesh has to seal against the glass wall.
Failure types seen here: completeness (a required part missing); part-order (components stacked or connected in the wrong sequence); physical-plausibility (something that couldn't be built or wouldn't work); 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 23 Sept, 3 rounds, $0.21The cleanest press on the bench, with the cross-plate on top of the mesh.
run 2 of 23 Sept, 2 rounds, $0.21The stack the right way up in two rounds, with the frame drawn beside the glass instead of under it.
run 3, 3 rounds, $0.22Right in round 1, then it swapped the plates, and its filter is wider than the glass.
run 1 of 22 Sept, 5 rounds, $0.36Seven parts, the stack the right way up, and labels on both sides of one axis.
run 2 of 22 Sept, 4 rounds, $0.37The same result in one round fewer, with a handle drawn on the carafe.
run 3, 2 rounds, $0.45Cross-plate on top of the mesh again, and plates that are the wrong diameter for the glass. Cut off by provider timeouts after round 2.
run 1, 2 rounds, $0.27The filter stack is in the right order but drawn under the carafe, which is not how it goes together.
run 2, 1 round, $0.18Right order in one round, but the plates are under 80% of the glass width, and the cross-plate is drawn behind the carafe rim.
run 3, 1 round, $0.21Right order in one round, but the plates and mesh are barely over half the width of the carafe they have to seal against.
run 1, 4 rounds, $0.02Three filter pieces labelled as one assembly, so which plate is which cannot be read, and all three are under 80% of the glass width.
run 2, 3 rounds, $0.01Calls its top piece a spiral cross plate and puts the mesh at the bottom, with the lid below the whole filter and every plate wider than the carafe.
run 3, 4 rounds, $0.02A plate missing, the cross-plate above the mesh, the lid below the filter, and plates wider than the carafe.
run 1, 3 rounds, $0.04No spiral plate and no separate knob, and the cross retainer sits above the mesh.
run 2, 2 rounds, $0.02A coil spring and a mesh, no cross-plate, the lid under the filter, and a mesh wider than the carafe.
run 3, 2 rounds, $0.03Mesh and a flange disc, no spiral plate or cross-plate. Cut off after round 2.Earlier results ran under an older prompt, so they are at the bottom, next to the question they were asked.
Older prompts
Prompt v1, retired 2026-09-20
Draw an exploded assembly diagram of a French press, as an SVG.
The ground truth only ever described the three-piece plunger, but the prompt never said which press to draw. A model that drew a two-piece travel press, where a retainer sits over the mesh and there is no spiral plate at all, would have failed the stack-order check for answering a question I had not asked. The five results below all ran under v1 and are scored against it.
