Kimi K3
Moonshot. Run in Kimi Code CLI, OpenCode, Raw API call.
106 runs on 90 prompts · 90% finished
Results
The best run of each of its 90 tests first, then every other run. Filter them in the explorer
95 finished runs
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live scene
Rendered video
Numbers: finish rate, cost, wall time, draft scores
- Runs
- 106
- on 90 prompts
- Finished
- 90%
- 95 of 106
- Median cost, finished run
- $1.28
- n=30
- Median wall time, finished run
- 11m 50s
- n=95
- Recorded spend
- $67.64
- $4.85 on unfinished runs; 73 without a recorded cost
From the numbers
Set against the other models with at least 3 runs (top or bottom third), and its draft-score ranks within single prompts. Nothing here is averaged across prompts.
Strengths
- Draft score 8.5/10 on Lighthouse on a headland at night (in OpenCode): ranked 2nd of 12
- Draft score 9.5/10 on Animated pelican riding a bicycle (SMIL/CSS) (in Kimi Code CLI): ranked 2nd of 12
Weaknesses
- 11m 50s (n=95): 18th of 23 models with at least 3 (median wall time of a finished run)
- Draft score 8/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (in Kimi Code CLI): ranked 9th of 12
- Draft score 7.5/10 on Lighthouse on a headland at night (in Kimi Code CLI): ranked 9th of 12
- 6 runs ended as the model's own failure (DNF, class MODEL)
- Other unfinished runs, not counted against the model: 3 INFRA, 2 TIMEOUT
Best and hardest run
Best run
Picked as its best draft-score rank on a prompt.
-
Kimi K3, Kimi Code CLI, Animated pelican riding a bicycle (SMIL/CSS)
Kimi Code CLI 2.1.0, max effort, via Moonshot
skills off
13m 32s
$0.74 estimated
OK
AI judge 9.5/10 Draft scorerubric
Hardest run
Picked as its costliest run that did not finish.
-
Did not finish · re-run queued
Kimi K3, Kimi Code CLI, gpu_fluid_sim
Kimi Code CLI 2.1.0, max effort, via Moonshot
skills off
Stopped by us
57m 02s $2.29 estimated
Draft scores by prompt
| Isometric SVG town with sprite sheet | Kimi Code CLI | 9/10 draft | 8th of 11 |
|---|---|---|---|
| Isometric SVG town with sprite sheet | OpenCode | 9/10 draft | 8th of 11 |
| Lighthouse on a headland at night | OpenCode | 8.5/10 draft | 2nd of 12 |
| Lighthouse on a headland at night | Kimi Code CLI | 7.5/10 draft | 9th of 12 |
| MacBook Pro 16" realistic 3D render (Playcode benchmark) | OpenCode | 8.5/10 draft | 6th of 12 |
| MacBook Pro 16" realistic 3D render (Playcode benchmark) | Kimi Code CLI | 8/10 draft | 9th of 12 |
| Pelican riding a bicycle (control) | OpenCode | 9/10 draft | 6th of 13 |
| Pelican riding a bicycle (control) | Kimi Code CLI | 8.5/10 draft | 9th of 13 |
| Animated pelican riding a bicycle (SMIL/CSS) | Kimi Code CLI | 9.5/10 draft | 2nd of 12 |
| Animated pelican riding a bicycle (SMIL/CSS) | OpenCode | 9/10 draft | 6th of 12 |
By harness
| Kimi Code CLI | 15 | 14 | 93% | $1.30 n=14 | 22m 50s n=14 |
|---|---|---|---|---|---|
| Raw API call | 71 | 65 | 92% | not recorded: no finished run with a recorded cost | 9m 10s n=65 |
| OpenCode | 20 | 16 | 80% | $1.20 n=16 | 27m 52s n=16 |
All 90 tests and experiment collections
- Animated pelican riding a bicycle (SMIL/CSS) (2 runs)
- Isometric SVG town with sprite sheet (2 runs)
- Pelican riding a bicycle (control) (2 runs)
- Lighthouse on a headland at night (2 runs)
- MacBook Pro 16" realistic 3D render (Playcode benchmark) (2 runs)
- Live watercolour: Venetian canal (2 runs)
- Chart morph (bars, line, donut) (1 run)
- 025 Gravity Well (1 run)
- 024 Tower Stack (1 run)
- 019 Boids Lagoon (1 run)
- 021 Rotating CSS Cube (1 run)
- 013 Soft Body Blobs (1 run)
- 016 Heartbeat Line (1 run)
- 009 Forest Fire Automaton (2 runs)
- 011 Pure CSS Sunset (1 run)
- 006 Rainfall Bars (1 run)
- 007 Kinetic Headline (1 run)
- 002 Pendulum Wave (1 run)
- Mobile touch arcade game (1 run)
- Glass launch UI (1 run)
- Sales Dashboard Card (1 run)
- Cycling Peloton (1 run)
- Cyberpunk Alley (1 run)
- Corgi Catching a Frisbee (1 run)
- Circuit Board Close-Up (1 run)
- Chalkboard Coffee Lettering (1 run)
- Top-Down City Park (1 run)
- Chameleon on a Branch (1 run)
- Chat Conversation UI (1 run)
- Canyon River (1 run)
- Container Ship (1 run)
- Campus Map (1 run)
- Butterfly Life Cycle (1 run)
- Lowpoly dragon (1 run)
- Calendar Month View (1 run)
- Bubble Tea (1 run)
- Exploded Burger (1 run)
- Boxing Ring (1 run)
- Breakfast Skillet (1 run)
- Labelled Bicycle Diagram (1 run)
- Bauhaus Composition (1 run)
- Basketball Slam Dunk (1 run)
- Battery Level Icons (1 run)
- Baseball Stadium (1 run)
- Barn Owl in Flight (1 run)
- Ballet Dancer (1 run)
- Fruit Sales Bar Chart (1 run)
- Autumn Forest Path (1 run)
- Bohr Model of an Atom (1 run)
- Stargazer with Telescope (1 run)
- Floating Astronaut (1 run)
- Art Deco Skyscraper (1 run)
- Ant Colony Cutaway (1 run)
- Flat App Icon Set (1 run)
- Alien Planet (1 run)
- Illustrated Alphabet A B C (1 run)
- Acoustic Guitar (1 run)
- Twin-Bell Alarm Clock (1 run)
- 017 Typewriter Poem (1 run)
- 023 Mortgage Calculator (1 run)
- 022 Mandala Painter (1 run)
- 020 Pricing Table Trio (1 run)
- 018 Step Sequencer Grid (1 run)
- 014 Snake Reimagined (1 run)
- 010 Toggle Switch Gallery (1 run)
- 015 Color Palette Studio (1 run)
- 012 Magnetic Filings (1 run)
- 008 Pocket Theremin (1 run)
- Sleeping Curled Cat (1 run)
- 003 Orbital Coffee Landing (1 run)
- 004 Brick Breaker Neon (1 run)
- 005 Focus Timer Orbit (1 run)
- Fjord scene (Three.js) (2 runs)
- Pelican car france (2 runs)
- Dolphin vlog bicycle (2 runs)
- Whale clouds (2 runs)
- California brown pelican riding a bicycle (detailed spec) (2 runs)
- BMW M4 Competition, side view (flat vector) (2 runs)
- Xbox 360 controller (2 runs)
- DualSense controller, max fidelity (2 runs)
- GPU fluid sim (1 run)
- Artwork style pass (1 run)
- Folding road (2 runs)
- Mechanical bee (1 run)
- Animated short (HTML/JS) (1 run)
- Ant vs gardener (1 run)
- Chef Flipping a Pan (1 run)
- Chessboard Mid-Game (1 run)
- 001 Ink Bloom Garden (1 run)
- X16 engine (1 run)
Experiment collections: HTML Files (37) · SVG (64) · Three.js (11) · Blender (1) · Harness lane (106)