OpenCode
12 models ran in it: Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, DeepSeek V4.1 Flash, GLM-5.3, GPT-6 Astra, Gemini 3.8 Flash, Grok 4.7, Kimi K3, LongCat 2.5 Preview, Muse Spark 1.3, Space Bunny Alpha.
307 runs on 110 prompts · 80% finished
Results
The best run of each of its 110 tests first, then every other run. Filter them in the explorer
247 finished runs
Live scene
Live scene
Live scene
Live scene
Live scene
Live scene
Live scene
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Live demo
Islamic star tiling (agent version)
LongCat 2.5 Preview
OpenCode 1.18.32
skills off
Failed · Stopped by us
City park top down (agent version)
LongCat 2.5 Preview
OpenCode 1.18.32
skills off
Failed · Stopped by us
- Post-run video; no posterPost-run video
AI dungeon
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
Rendered by the site after the unfinished run
How internet works
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Infrastructure failure
Video not rendered
Bacterium to quantum zoom
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Infrastructure failure
Project source only; video not rendered
Quadruped rig walk
DeepSeek V4.1 Flash
OpenCode 1.18.32
skills off
Failed · Error
3D model exported by the site · Explore 3D model
DualSense controller, max fidelity
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
Isometric SVG town with sprite sheet
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
Isometric SVG town with sprite sheet
Claude Opus 5.5
OpenCode 1.18.32
skills off
Failed · Did not finish
MacBook Pro 16" realistic 3D render (Playcode benchmark)
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
MacBook Pro 16" realistic 3D render (Playcode benchmark)
Claude Opus 5.5
OpenCode 1.18.32
skills off
Failed · Stopped by us
Lighthouse on a headland at night
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
X16 engine
Claude Opus 5.5
OpenCode 1.18.32
skills off
Failed · Stopped by us
Blank until you tap or click it (the demo waits for input).
Video printing press collage hyperframes
GLM-5.3
OpenCode 1.18.32
skills off
Failed · Stopped by us
No rendered video or project published
Video printing press collage hyperframes
DeepSeek V4.1 Flash
OpenCode 1.18.32
skills off
Failed · Error
No rendered video or project published
Video printing press collage remotion
DeepSeek V4.1 Flash
OpenCode 1.18.32
skills off
Failed · Stopped by us
No rendered video or project published
Folding road
DeepSeek V4.1 Flash
OpenCode 1.18.32
skills off
Failed · Stopped by us
The run did not publish main.js; it cannot run as built without them, so it is not played here.
Space bunny
Claude Opus 5.5
OpenCode 1.18.32
skills off
Failed · Stopped by us
The model's code errors on load: Identifier 'k' has already been declared
Glass launch UI
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Infrastructure failure
Video not rendered
California brown pelican riding a bicycle (detailed spec)
Claude Sonnet 5.5
OpenCode 1.18.33
skills off
Failed · Stopped by our timeout
GPU fluid sim
LongCat 2.5 Preview
OpenCode 1.18.32
skills off
Failed · Stopped by us
The model's code errors on load: ambientTime is not defined
Numbers: finish rate, cost, wall time, draft scores
- Runs
- 307
- on 110 prompts
- Finished
- 80%
- 247 of 307
- Median cost, finished run
- $1.06
- n=246
- Median wall time, finished run
- 19m 49s
- n=247
- Recorded spend
- $1,281.79
- $240.23 on unfinished runs; 7 without a recorded cost
From the numbers
Set against the other harnesses with at least 3 runs (top or bottom third), and its draft-score ranks within single prompts. Nothing here is averaged across prompts.
Strengths
- $1.06 (n=246): 4th of 12 harnesses with at least 3 (median cost of a finished run)
- Draft score 9.5/10 on Pelican riding a bicycle (control) (Claude Fable 5.1): ranked 1st of 13
- Draft score 9.5/10 on Pelican riding a bicycle (control) (GLM-5.3): ranked 1st of 13
- Draft score 9/10 on Lighthouse on a headland at night (Claude Fable 5.1): ranked 1st of 12
- Draft score 10/10 on Isometric SVG town with sprite sheet (Claude Fable 5.1): ranked 1st of 11
- Draft score 9/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (Claude Fable 5.1): ranked 2nd of 12
- Draft score 9/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (GLM-5.3): ranked 2nd of 12
- Draft score 9/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (GPT-6 Astra): ranked 2nd of 12
- Draft score 8.5/10 on Lighthouse on a headland at night (Claude Opus 5.5): ranked 2nd of 12
- Draft score 8.5/10 on Lighthouse on a headland at night (Kimi K3): ranked 2nd of 12
- Draft score 9.5/10 on Animated pelican riding a bicycle (SMIL/CSS) (Claude Fable 5.1): ranked 2nd of 12
- Draft score 9.5/10 on Animated pelican riding a bicycle (SMIL/CSS) (Claude Opus 5.5): ranked 2nd of 12
- Draft score 9.5/10 on Animated pelican riding a bicycle (SMIL/CSS) (GPT-6 Astra): ranked 2nd of 12
- Draft score 9.5/10 on Isometric SVG town with sprite sheet (GLM-5.3): ranked 2nd of 11
- Draft score 9.5/10 on Isometric SVG town with sprite sheet (GPT-6 Astra): ranked 2nd of 11
Weaknesses
- Draft score 7.5/10 on Lighthouse on a headland at night (Grok 4.7): ranked 9th of 12
- Draft score 7.5/10 on Pelican riding a bicycle (control) (Grok 4.7): ranked 11th of 13
- Draft score 6.5/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (DeepSeek V4.1 Flash): ranked 11th of 12
- Draft score 6.5/10 on Lighthouse on a headland at night (DeepSeek V4.1 Flash): ranked 11th of 12
- Draft score 8.5/10 on Animated pelican riding a bicycle (SMIL/CSS) (Grok 4.7): ranked 11th of 12
- Draft score 6.5/10 on Pelican riding a bicycle (control) (DeepSeek V4.1 Flash): ranked 13th of 13
- Draft score 4.5/10 on MacBook Pro 16" realistic 3D render (Playcode benchmark) (Grok 4.7): ranked 12th of 12
- Draft score 5.5/10 on Lighthouse on a headland at night (GLM-5.3): ranked 12th of 12
- Draft score 8/10 on Animated pelican riding a bicycle (SMIL/CSS) (DeepSeek V4.1 Flash): ranked 12th of 12
- 2 runs ended as the model's own failure (DNF, class MODEL)
- Other unfinished runs, not counted against the model: 2 HARNESS, 13 INFRA, 42 TIMEOUT, 1 STOPPED
Best and hardest run
Best run
Picked as its best draft-score rank on a prompt.
-
GLM-5.3, OpenCode, Pelican riding a bicycle (control)
OpenCode 1.18.32, max effort, via Fireworks
skills off
9m 06s
$0.33 reported by OpenCode est. $0.33
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
Claude Fable 5.1, OpenCode, threejs_jellyfish
OpenCode 1.18.32, max effort, via Anthropic
skills off
Stopped by us
42m 58s $14.23 reported by OpenCode est. $14.23
Draft scores by prompt
Show the table (33 rows)
By model
| Grok 4.7 | 6 | 6 | 100% | $7.65 n=6 | 31m 01s n=6 |
|---|---|---|---|---|---|
| GPT-6 Astra | 50 | 49 | 98% | $4.30 n=49 | 18m 28s n=49 |
| Space Bunny Alpha | 13 | 12 | 92% | $0 n=11 | 7m 19s n=12 |
| Muse Spark 1.3 | 12 | 11 | 92% | $0 n=11 | 5m 38s n=11 |
| DeepSeek V4.1 Flash | 47 | 42 | 89% | $0.13 n=42 | 19m 07s n=42 |
| LongCat 2.5 Preview | 52 | 42 | 81% | $0 n=42 | 9m 18s n=42 |
| Kimi K3 | 20 | 16 | 80% | $1.20 n=16 | 27m 52s n=16 |
| Claude Fable 5.1 | 30 | 23 | 77% | $8.67 n=23 | 28m 33s n=23 |
| Claude Opus 5.5 | 34 | 25 | 74% | $15.39 n=25 | 1h 04m n=25 |
| GLM-5.3 | 24 | 17 | 71% | $1.08 n=17 | 26m 23s n=17 |
| Claude Sonnet 5.5 | 18 | 4 | 22% | $6.56 n=4 | 37m 09s n=4 |
| Gemini 3.8 Flash | 1 | 0 | 0% | not recorded: no finished run with a recorded cost | not recorded: no finished run with a recorded time |
All 110 tests
- DualSense controller, max fidelity (11 runs)
- Isometric SVG town with sprite sheet (12 runs)
- MacBook Pro 16" realistic 3D render (Playcode benchmark) (12 runs)
- Animated pelican riding a bicycle (SMIL/CSS) (10 runs)
- Pelican riding a bicycle (control) (11 runs)
- Lighthouse on a headland at night (10 runs)
- Dragon ship (2 runs)
- Tidewater (7 runs)
- X16 engine (9 runs)
- Fjord scene (Three.js) (6 runs)
- Ant vs gardener (7 runs)
- Momijigari (6 runs)
- Mobile touch arcade game (2 runs)
- Power grid day (1 run)
- World weather stations (1 run)
- Marathon results (1 run)
- Football league season (1 run)
- Maze generate and solve (1 run)
- Saas growth metrics (1 run)
- Coffee chain sales (1 run)
- Ecommerce funnel (1 run)
- Drum machine daw (2 runs)
- Hash table showdown (1 run)
- Raft consensus (2 runs)
- GPU boids 50k (2 runs)
- Paint studio (2 runs)
- Code editor (2 runs)
- Kanban board (1 run)
- Balanced trees (1 run)
- Calendar planner (2 runs)
- Photo editor (2 runs)
- Mst kruskal prim (1 run)
- Team chat (1 run)
- Edit distance dp (1 run)
- Spreadsheet (2 runs)
- Retro desktop (2 runs)
- Sorting race (2 runs)
- Pathfinding race (2 runs)
- Stillwater pond (2 runs)
- Humanoid walk cycle (2 runs)
- Video printing press collage hyperframes (3 runs)
- Video printing press collage remotion (2 runs)
- Humanoid motion cycle (2 runs)
- Folding road (9 runs)
- Animated short (HTML/JS) (4 runs)
- Creature robot blueprint (2 runs)
- Spongebob (2 runs)
- Jellyfish (6 runs)
- Space bunny (5 runs)
- Pixel lofi loop (2 runs)
- BMW M4 Competition, side view (flat vector) (10 runs)
- Xbox 360 controller (9 runs)
- Artemis II (4 runs)
- Glass launch UI (2 runs)
- Soccer bicycle kick (agent version) (1 run)
- Treasure map (agent version) (1 run)
- Lowpoly dragon (3 runs)
- Voronoi cells (agent version) (1 run)
- Human heart anatomy (agent version) (1 run)
- Human skeleton front (agent version) (1 run)
- Food pyramid (agent version) (1 run)
- Grand piano (agent version) (1 run)
- Logo sushi bar (agent version) (1 run)
- Hand five fruits (agent version) (1 run)
- Vinyl turntable (agent version) (1 run)
- Boxing ring (agent version) (1 run)
- Water cycle diagram (agent version) (1 run)
- Logo coffee roaster (agent version) (1 run)
- Hello neon sign (agent version) (1 run)
- Halloween drippy boo (agent version) (1 run)
- Dragon sleeping castle (agent version) (1 run)
- Detective noir (agent version) (1 run)
- Japanese pagoda (agent version) (1 run)
- Chat interface (agent version) (1 run)
- Suspension bridge fog (agent version) (1 run)
- Astronaut floating (agent version) (1 run)
- Sewing machine (agent version) (1 run)
- Weather widget (agent version) (1 run)
- Latte art (agent version) (1 run)
- Swiss army knife (agent version) (1 run)
- Mountain lake dawn (agent version) (1 run)
- Rice terraces (agent version) (1 run)
- Sushi platter (agent version) (1 run)
- Retro crt computer (agent version) (1 run)
- Film camera (agent version) (1 run)
- Steam locomotive (agent version) (1 run)
- Hummingbird flower (agent version) (1 run)
- Ladybug leaf (agent version) (1 run)
- Cargo ship containers (agent version) (1 run)
- Panda bamboo (agent version) (1 run)
- Red fox snow (agent version) (1 run)
- Dolphin vlog bicycle (9 runs)
- California brown pelican riding a bicycle (detailed spec) (10 runs)
- Pelican car france (9 runs)
- Whale clouds (9 runs)
- GPU fluid sim (3 runs)
- Eiffel (4 runs)
- Procedural armor suit (2 runs)
- Slime softbody (2 runs)
- Game battleship (2 runs)
- Jack o lanterns (agent version) (1 run)
- Islamic star tiling (agent version) (1 run)
- City park top down (agent version) (1 run)
- Cyberpunk alley (agent version) (1 run)
- AI dungeon (1 run)
- How internet works (1 run)
- Bacterium to quantum zoom (1 run)
- Quadruped rig walk (1 run)