How does Kimi K3.1 Pro do on the five busiest tasks?

In short

It built the hopper right every time.

Finding

Kimi K3.1 Pro went up on OpenRouter on 23 October. I ran a couple of tasks.

Run 1 Run 2 Run 3
French press right way up upside down mesh at the bottom
AeroPress filter too small filter too small filter too small
Bean cosy bean too long right right
Grinder hopper built right built right built right
Pruned shrub flat side flat side, no shrub flat side, no shrub
SVG: exploded French press, by kimi-k3.1-pro
Run 1. Spiral plate, mesh, cross plate, which is right.$0.18
SVG: exploded French press, by kimi-k3.1-pro
Run 2. The same three upside down.$0.18

The press filters consistently don’t fit the glass for Kimi K3.1, though.

L-system: hedge-pruned coffee shrub, by kimi-k3.1-pro
Run 1. The left side cut flat by the rules.$0.22
L-system: hedge-pruned coffee shrub, by kimi-k3.1-pro
Run 3. Still flat on the right, but it no longer reads as a shrub.$0.26

Kimi is the second model after Pareto 26.9 to build the hopper right in all three runs, so the hopper is now retired from the board.

The 15 runs cost $3.15.

By email

How it failed

false, incomplete, unreadable

Older, 20 Oct

How good are models at fixing things?

Newer, 8 Nov

Do models say no when a set of coffee order is impossible?