Get the Agent OS + join 4,000+ founders inside the AI Profit Boardroom → Join AIPB ($59/mo)
Game

Doom

Doom — put monsters in the raycaster maze and let them chase you.

CategoryGame
Models tested27
Scored22/27
Avg score7.43/10
WinnerHermes MoA

What I asked each model — the Doom prompt

Every model on this page got this exact prompt inside the Agent Operating System: Doom — put monsters in the raycaster maze and let them chase you.

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. 27 frontier models have attempted it so far: Claude Fable 5, Fugu Ultra, Fugu Ultra 1.1, Fugu Mini, Fusion, Gemini 3.6 Flash, GLM-5.2, GPT-5.6 Sol, Grok, Grok 4.6, Hy3, 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, DeepSeek V4 Pro, DeepSeek V4 Flash, Kimi K2.7 · Fast, Kimi K2.7 · No-Think, Kimi K2.7 · Quality.

Why this task matters. Doom 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 Doom

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
• 8.1/10

What I saw: Iterated rebuild lands a complete DOOM-style raycaster: textured corridor walls with correct perspective, a gun and crosshair, a visible demon enemy, an HP/kills/wave HUD and a live minimap tracking monsters. Move, turn, strafe and shoot all respond (verified) — a convincing, playable first impression.

▶ Play Claude Fable 5's attempt →
Fugu Ultra Sakana AI
• 7.0/10

What I saw: Ultra v2 (gap-fill, 16-min direct call) — 45KB Doom-style raycaster FPS: sprite enemies, gun + muzzle flash, ammo/health HUD, 2 rAF loops, 9 input handlers. Smoke-test STATIC because movement is gated on pointer-lock (update() returns while the start-overlay is open) and headless Chrome can't engage pointer-lock — same limitation as skyrim/crypt. Renders a full 274KB scene; flagged for manual verification.

▶ Play Fugu Ultra's attempt →
Fugu Ultra 1.1 Sakana AI
• 8.4/10 · 3D demon shooter

What I saw: Strong atmospheric 3D maze with a visible horned demon enemy, working shotgun, minimap with tracked enemies, and active combat ('CLAWED' hit feedback, HP dropped to 082) — polished HUD and lighting. Falls just shy of the field's best; enemies are more Three.js models than true raycaster sprites, but combat clearly works.

▶ Play Fugu Ultra 1.1's attempt →
Fugu Mini Sakana AI
• 8.0/10

What I saw: Doom-style raycaster FPS with sprite enemies. Smoke-test PASS (pixel change after click+keys).

▶ Play Fugu Mini's attempt →
Fusion OpenRouter
🥈 8.5/10

What I saw: Doom-style raycaster shooter: sprite enemies, gun with muzzle flash, ammo + health HUD. WASD + mouse-look + click to fire. 20KB of game logic.

▶ Play Fusion's attempt →
• 6.8/10

What I saw: Strong atmospheric raycaster-style scene with polished textured walls, weapon model, and a full functional HUD (health/stamina/ammo/kills), but the only visible 'enemy' is a plain cyan cube with no demon presence or visible combat/chase happening in the frame, undercutting the core Doom brief.

▶ Play Gemini 3.6 Flash's attempt →
GLM-5.2 Zhipu / Z.ai
• 8.0/10

What I saw: All three are real, playable shooters. Opus drops you in a corridor with an imp dead ahead — gun, crosshair and HUD framed like a screenshot. Kimi matches it: a monster down a textured hall, health, ammo, minimap. GLM ships a gorgeous 'HAZARD PROTOCOL' title screen with a working game behind it, though it too spawns facing a wall. Opus by a hair on the cleanest fight.

▶ Play GLM-5.2's attempt →
GPT-5.6 Sol OpenAI
• 8.4/10 · polished demon raycaster

What I saw: Clean raycaster with atmospheric red-lit corridors, a well-drawn menacing demon sprite with glowing eyes and teeth, weapon viewmodel, minimap with hostile dots, and a cohesive DOOM HUD; slightly below the top only for the somewhat cartoonish monster and gradient walls that read more stylized than gritty, but it clearly nails the brief and is shippable.

▶ Play GPT-5.6 Sol's attempt →
Grok xAI
🥈 8.5/10

What I saw: Doom-style FPS with sprite enemies, gun + muzzle flash + ammo/health HUD, textures, pointer-lock mouse-look. 22KB.

▶ Play Grok's attempt →
Grok 4.6 xAI
• 6.4/10

What I saw: Re-scored after a real-GPU playtest. It LOOKS like a top-tier Doom - a red demon filling the screen, a shotgun sprite, a working minimap with enemy dots, and ammo that decrements on fire. But at a real 70fps each demon calls hurtPlayer(12) on a 0.18s cooldown, about 67 damage per second each, so I died 8 times in 6 seconds and never got a fight. My first pass only scored it 8.4 because the headless renderer ran at 2fps and I never reached contact range - the slow harness hid a hard balance failure. Handed Grok 4.6 that exact evidence and it rebalanced its own combat: the live demo now survives a 20-second fight with kills on the board and HP draining gradually.

▶ Play Grok 4.6's attempt →

The winner on Doom

Hermes MoA took gold on this task.

What I saw: A complete, polished raycaster that nails the Doom screenshot framing — corridor with imps dead ahead, detailed canvas-drawn imp sprites with bob/flash/death states, muzzle flash, screen-shake kick, hit-scan with line-of-sight checks, and a clean DOOM-branded HUD with health/ammo/kills. The DDA casting, z-buffered sprite sorting, brick shading, and damage vignette edge it slightly past SOLO Opus 4.8 and Fusion (8.5) on visual cohesion and code quality, though it lacks pointer-lock mouse-look (drag-turn only).

See Hermes MoA's full model card: /models/moa. Direct head-to-head against the runner-up: Hermes MoA vs Fusion.

Every attempt — live, playable

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

How I scored Doom — 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.

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