What's the prompt for the Twilightvale test?

Twilight Vale — 3D open-world RPG with combat, terrain, weather. Every model receives this exact prompt, one shot, single HTML file out.

Game

Twilightvale

Q: What's the best AI model for Twilightvale?

Fusion — RETRY @ 24K tokens — now complete: 44KB three.js + WebGL using renderer.setAnimationLoop (three.js native loop), 6 input handlers, full update() per-frame: player + NPCs + enemies + weather + day/night + HUD. Densest build on the bench.

Q: How many AI models attempted Twilightvale?

24 models on Goldie Bench have attempted Twilightvale: Claude Fable 5, Fugu Ultra, Fugu Ultra 1.1, Fusion, Gemini 3.6 Flash, GLM-5.2, GPT-5.6 Sol, Grok, Inkling, Kimi K2.7, Kimi K3, MiniMax M3, Hermes MoA, Muse Spark 1.2, 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.

Twilight Vale — 3D open-world RPG with combat, terrain, weather.

CategoryGame

Models tested24

Scored19/24

Avg score7.78/10

WinnerFusion

What I asked each model — the Twilightvale prompt

Every model on this page got this exact prompt inside the Agent Operating System: Twilight Vale — 3D open-world RPG with combat, terrain, weather.

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. 24 frontier models have attempted it so far: Claude Fable 5, Fugu Ultra, Fugu Ultra 1.1, Fusion, Gemini 3.6 Flash, GLM-5.2, GPT-5.6 Sol, Grok, Inkling, Kimi K2.7, Kimi K3, MiniMax M3, Hermes MoA, Muse Spark 1.2, 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. Twilightvale is a textbook test of game-class capability — the kind of build that exposes whether a model is doing pattern-matching or actual reasoning. A model that ships this in one shot is usually safe to wire into your agent loop for harder tasks of the same shape.

How each model handled Twilightvale

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.

Claude Fable 5 Anthropic

🥉 9.0/10

What I saw: AAA rebuild is showcase-grade: an authored hooded hero with a glowing trailed weapon on a moody twilight field of glowing mushrooms, floating crystals and rune-stones, live storm weather with lightning, collectible shards, and a cohesive HUD (vitality, wave/kills/wisps, weather, compass, ATK). Move, orbit and attack verified. Dense, atmospheric and clearly readable.

▶ Play Claude Fable 5's attempt →

Fugu Ultra Sakana AI

🥉 9.0/10 · winner · open-world depth

What I saw: Ultra v2 — 61.8KB open-world RPG (village, NPCs, weather, day/night). Smoke-test PASS. Densest Ultra build on the bench.

▶ Play Fugu Ultra's attempt →

Fugu Ultra 1.1 Sakana AI

• 8.6/10 · Atmospheric RPG World

What I saw: Gorgeous cohesive twilight scene with detailed hero holding a weapon, visible enemies (creature + humanoid), wooden bridge, lampposts, cottage, river, pickups and a working minimap/HUD with quest and kill tracker. Strong art direction and populated world with combat framing; only minor risk is verifying enemy AI aggression, but the visible enemies plus polish put it at the top of the field.

▶ Play Fugu Ultra 1.1's attempt →

Fusion OpenRouter

🥇 9.5/10 · winner · open world RPG

What I saw: RETRY @ 24K tokens — now complete: 44KB three.js + WebGL using renderer.setAnimationLoop (three.js native loop), 6 input handlers, full update() per-frame: player + NPCs + enemies + weather + day/night + HUD. Densest build on the bench.

▶ Play Fusion's attempt →

Gemini 3.6 Flash Google

• 7.8/10

What I saw: Strong atmospheric twilight scene with cohesive village, terrain, fog and a well-armed character plus polished HUD, minimap and quest tracker; however no enemies are visible in the shot (minimap shows red dots but combat isn't demonstrated), keeping it just short of the top tier.

▶ Play Gemini 3.6 Flash's attempt →

GLM-5.2 Zhipu / Z.ai

• 7.5/10

What I saw: 54KB · plays clean · plain

▶ Play GLM-5.2's attempt →

GPT-5.6 Sol OpenAI

• 7.8/10

What I saw: Renders cleanly with a cohesive low-poly scene—hooded hero with sword, shrine with pillars/orb, wraith enemies, forest, and a polished HUD with quest/stats/weather. Solid and shippable, but the pastel fog washes out the promised 'twilight' mood and the terrain looks flat/snowy rather than a dramatic vale, keeping it short of the field's best.

▶ Play GPT-5.6 Sol's attempt →

Grok xAI

🥇 9.5/10 · winner · open world depth

What I saw: Twilight Vale — 3D open-world RPG with hand-crafted village, NPCs, combat, day/night, weather, inventory. 38KB — densest build of the bench, edges out Fusion's 32KB.

▶ Play Grok's attempt →

Inkling Thinking Machines

• 4.5/10

What I saw: Clean UI and title, but the 3D scene barely renders — the terrain/ground plane appears invisible or floated away, leaving just a floating sphere and stray tree trunks against a bare background, so it reads broken rather than an open-world realm. Nice HUD polish and combat-log styling can't rescue the failed terrain/world presentation.

▶ Play Inkling's attempt →

Kimi K2.7 Moonshot AI

• 8.5/10

What I saw: 64KB open-world RPG with village, NPCs, combat, day/night cycle. Densest Kimi build.

▶ Play Kimi K2.7's attempt →

The winner on Twilightvale

Fusion took gold on this task. winner · open world RPG.

See Fusion's full model card: /models/fusion.

Every attempt — live, playable

Side by side. Click any tile to run that model's actual one-shot HTML in a new tab.

▶ LIVE

Claude Fable 5 🥉

Anthropic

AAA rebuild is showcase-grade: an authored hooded hero with a glowing trailed weapon on a moody twilight field of glowing mushrooms, floating crystals and rune-stones, live storm weather with lightning, collectible shards, and a cohesive HUD (vitality, wave/kills/wisps, weather, compass, ATK). Move, orbit and attack verified. Dense, atmospheric and clearly readable.

▶ LIVE

Fugu Ultra 🥉

Sakana AI

Ultra v2 — 61.8KB open-world RPG (village, NPCs, weather, day/night). Smoke-test PASS. Densest Ultra build on the bench.

▶ LIVE

Fugu Ultra 1.1

Sakana AI

Gorgeous cohesive twilight scene with detailed hero holding a weapon, visible enemies (creature + humanoid), wooden bridge, lampposts, cottage, river, pickups and a working minimap/HUD with quest and kill tracker. Strong art direction and populated world with combat framing; only minor risk is verifying enemy AI aggression, but the visible enemies plus polish put it at the top of the field.

▶ LIVE

Fusion 🥇

OpenRouter

RETRY @ 24K tokens — now complete: 44KB three.js + WebGL using renderer.setAnimationLoop (three.js native loop), 6 input handlers, full update() per-frame: player + NPCs + enemies + weather + day/night + HUD. Densest build on the bench.

▶ LIVE

Gemini 3.6 Flash

Google

Strong atmospheric twilight scene with cohesive village, terrain, fog and a well-armed character plus polished HUD, minimap and quest tracker; however no enemies are visible in the shot (minimap shows red dots but combat isn't demonstrated), keeping it just short of the top tier.

▶ LIVE

GLM-5.2

Zhipu / Z.ai

54KB · plays clean · plain

▶ LIVE

GPT-5.6 Sol

OpenAI

Renders cleanly with a cohesive low-poly scene—hooded hero with sword, shrine with pillars/orb, wraith enemies, forest, and a polished HUD with quest/stats/weather. Solid and shippable, but the pastel fog washes out the promised 'twilight' mood and the terrain looks flat/snowy rather than a dramatic vale, keeping it short of the field's best.

▶ LIVE

Grok 🥇

xAI

Twilight Vale — 3D open-world RPG with hand-crafted village, NPCs, combat, day/night, weather, inventory. 38KB — densest build of the bench, edges out Fusion's 32KB.

▶ LIVE

Inkling

Thinking Machines

Clean UI and title, but the 3D scene barely renders — the terrain/ground plane appears invisible or floated away, leaving just a floating sphere and stray tree trunks against a bare background, so it reads broken rather than an open-world realm. Nice HUD polish and combat-log styling can't rescue the failed terrain/world presentation.

▶ LIVE

Kimi K2.7

Moonshot AI

64KB open-world RPG with village, NPCs, combat, day/night cycle. Densest Kimi build.

▶ LIVE

Kimi K3

Moonshot AI

Renders a coherent twilight low-poly world with cohesive purple/gold theming, full HUD (minimap, XP/HP/stamina bars, quest chip, NPC dialogue), trees, houses and water — clearly on-brief and shippable. However the camera clips badly into an oversized foreground terrain spike that dominates the frame, and the flat untextured cube houses look crude, keeping it below the polished top tier.

▶ LIVE

MiniMax M3 🥉

MiniMax

47KB — densest open-world. Village, NPCs, combat, day/night, weather, inventory.

▶ LIVE

Hermes MoA

Hermes · Mixture of Agents

Polished single-file three.js RPG with instanced trees/rocks, procedural terrain+river, weather/day-night, enemies with health bars, slash combat, and mobile controls — clearly beats SOLO Opus 4.8 (7.5) and is shippable. But it's lighter on open-world density than the leaders (Grok 9.5, Fusion 9.5, Fugu 9.0), so it strongly ties/slightly beats Opus rather than topping the field.

▶ LIVE

Muse Spark 1.2

How I scored Twilightvale — 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.

More game benchmarks: all tasks in the Game category · See the best AI model for Twilightvale · Back to the leaderboard

The same stack Julian uses

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.

4,000+founders

258documented wins

38countries

$59/momonthly

Join AIPB · $59/mo → Read the Agent OS guides →

Twilightvale

What I asked each model — the Twilightvale prompt

How each model handled Twilightvale

The winner on Twilightvale

Every attempt — live, playable

How I scored Twilightvale — methodology

Related

Run this stack yourself.

Join 4,000+ founders building with this stack.