Voxel
Voxel — voxel-art landscape (Minecraft-style).
What I asked each model — the Voxel prompt
Every model on this page got this exact prompt inside the Agent Operating System: Voxel — voxel-art landscape (Minecraft-style).
Single HTML file out. No iteration. No examples in the system prompt. Whatever each model produced on the first run is what's on this page. 23 frontier models have attempted it so far: Claude Fable 5, Fugu Ultra, Fugu Mini, Fusion, Gemini 3.6 Flash, GLM-5.2, GPT-5.6 Sol, Grok, Inkling, Kimi K2.7, Kimi K3, MiniMax M3, Hermes MoA, Opus 4.8, Claude Opus 5, Qwen 3.8, Qwen 3.7, Claude Sonnet 5, Kimi K2.7 · Fast, Kimi K2.7 · No-Think, Kimi K2.7 · Quality, DeepSeek V4 Pro, DeepSeek V4 Flash.
Why this task matters. Voxel is a textbook test of visual-class capability — the kind of build that exposes whether a model is doing pattern-matching or actual reasoning. Shipping this cleanly is the floor for what I expect from a frontier model — every model on the leaderboard should at least attempt it.
How each model handled Voxel
Ranked by my 0–10 score from the source comparison guides on agentos.guide. Click any to play the actual one-shot HTML the model produced.
What I saw: Iterated rebuild holds its excellent standard: a gorgeous floating voxel island with grass/stone/snow biomes, trees, ponds, drifting clouds and colorful flying birds, a day/night toggle and a bird counter. Orbit/zoom/glide controls respond (verified). Terrain reads a touch flat, keeping it just shy of the very top.
What I saw: Ultra v2 — Temple-Run voxel runner. Smoke-test PASS with 32.7% pixel diff — the single most reactive build. Replaces the earlier truncated voxel-fugu that was deleted.
What I saw: Mini gap-fill (round 2) — Temple-Run voxel runner. Smoke-test PASS with 17.0% pixel diff — works this time (the earlier Mini voxel was STATIC and got deleted).
What I saw: Closest thing to a real Temple Run any model has shipped: 3-lane runner with chunk streaming, jump + slide mechanics, coins, hurdles, gates, increasing speed, score/coins/speed/best HUD pills, touch-swipe support, gradient-text overlay card. Other voxel attempts were visuals only — this one's playable front to back.
What I saw: Strong polished UI and clean voxel intent, but the render is broken: the landscape is a flat cyan-water slab with no visible terrain, grass, or trees — only floating white cloud/snow blocks, so it reads as a washed-out empty plane rather than a Minecraft-style landscape.
What I saw: GLM built the densest, most detailed city — windowed skyscrapers, a speed + coins HUD. Opus ran the furthest with the cleanest motion (Score 303). Kimi's runner plays fine but is unforgiving — it crashes within seconds.
What I saw: The HUD, water plane, and clouds render cleanly with polished branding, but the entire voxel island terrain is rendered as an unlit black silhouette — likely a lighting/vertexColor/instanceColor failure — which guts the core Minecraft-style landscape brief.
What I saw: A colourful 3D voxel city with a score and coins HUD and a polished game-over card. Like every runner it ends fast — but the build is excellent.
What I saw: Renders a clean 3D voxel terrain with decent lighting, shadows, and polished UI overlay, but the random per-cube color scattering reads as noisy rather than coherent Minecraft-style biomes (grass/dirt/stone layers), and the drag-rotate control is disconnected from the actual auto-orbiting camera—solid but generic.
What I saw: GLM built the densest, most detailed city — windowed skyscrapers, a speed + coins HUD. Opus ran the furthest with the cleanest motion (Score 303). Kimi's runner plays fine but is unforgiving — it crashes within seconds.
The winner on Voxel
Fugu Ultra took gold on this task. winner · voxel runner.
What I saw: Ultra v2 — Temple-Run voxel runner. Smoke-test PASS with 32.7% pixel diff — the single most reactive build. Replaces the earlier truncated voxel-fugu that was deleted.
See Fugu Ultra's full model card: /models/fugu.
Every attempt — live, playable
Side by side. Click any tile to run that model's actual one-shot HTML in a new tab.
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVE
▶ LIVEHow I scored Voxel — methodology
Three axes, 0–10 each, averaged. Runs: drop the .html in a browser; if it opens to a broken page, it scores zero. Hits the brief: did the model ship the thing the prompt asked for, or a different thing it found easier. Looks good: visual polish, motion, interactivity — where most of the gap between gold and silver lives.
My scores trace back to the source comparison guides on agentos.guide. See the full methodology page for data provenance, including which source guide each cell's score came from.
Related
More visual benchmarks: all tasks in the Visual category · See the best AI model for Voxel · Back to the leaderboard
Run this stack yourself.
Every demo on this bench was built inside the Agent Operating System — one prompt, one shot, single HTML file out. The Agent OS, the prompts, the templates, the weekly walkthroughs and 4,000+ founders shipping with it every day all live inside the AI Profit Boardroom.