Added Moonshot Kimi K3
Moonshot shipped Kimi K3, a million-token model, so we added it to the matrix and ran all 34 one-shot prompts against it — same harness, same single-shot rules as every other build in the gallery.
It held up well: 18 of the 34 builds landed at four stars or better, and K3 was the only model in the set to earn a clean five on fluid-simulation — a real-time WebGL solver in one file, the hardest prompt we run. Where it slipped was games: a blank three.js cube from a stray JS error, a maze that never generates, and a couple of arcade builds that boot straight to a game-over screen.
The top builds below are its best-graded outputs; open any one for the live artifact, the full transcript, and the two-tier evaluation behind the score.
Top builds in this update
CLEAN “Colorful, swirling dye fluid simulation reacts vividly to click, drag, wheel, and even keyboard input.”— vision grade, anthropic/claude-sonnet-5
Full run + transcript →
CLEAN “Boids flock smoothly with glowing triangle shapes, flee from cursor circle, click adds boids; arrows/WASD have no effect.”— vision grade, anthropic/claude-sonnet-5
Full run + transcript →
CLEAN “Fireworks auto-launch and explode with colorful gravity-affected particles; clicks and other inputs also trigger launches.”— vision grade, anthropic/claude-sonnet-5
Full run + transcript →
CLEAN “Double pendulum simulation animates chaotically with colorful fading trail; enter/space seems to reset it.”— vision grade, anthropic/claude-sonnet-5
Full run + transcript →