Skip to builds

Counter-Strike 2 clone (Sol comparison): Fable 5 vs Opus 4.8

From “GPT 5.6 Sol: No-Hype Full Review & Testing”

  • Fable 5
  • Opus 4.8
  • GPT-5.6 Sol
  • GPT-5.5

The prompt

/goal Build me a visually stunning, genuinely playable browser clone of Counter-Strike 2 — a single bomb-defusal map in the spirit of Dust 2. Do NOT fork an existing project and ship it: if you use an open-source repo, use it only as a scaffold for core gameplay/mechanics, then build far better graphics and deeper gameplay on top. What ships must clearly surpass anything you started from.

Run this as a dynamic workflow in three phases:

1. RESEARCH — before writing any code, spawn parallel research sub-agents to study, from real reference online:
   (a) real Counter-Strike 2 / Dust 2 — the desert-tactical art style, the HUD (health, ammo, money, round timer, minimap), weapon feel, and the plant/defuse round loop. Find real Counter-Strike 2 gameplay screenshots from the web (gameplay videos, the CS wiki, Steam) and save them as your visual reference target;
   (b) browser FPS technique — Three.js first-person controller, hitscan shooting, bot AI, and modern WebGL lighting + post-processing (PBR, real-time shadows, ambient occlusion, tone mapping);
   (c) whether any open-source FPS offers usable gameplay scaffolding or asset geometry worth starting from — if so, take only the useful pieces and note the license to credit. Never ship a fork as-is.
   Write findings + the reference screenshots + any pulled scaffold to shared notes the build phase reads.

2. BUILD — fan across parallel sub-agents: the Dust 2-style map (bombsites, mid, crates, tight sightlines), first-person controller, weapons + hitscan shooting, bot AI (patrol + engage), round logic (buy/plant/defuse/win-lose), HUD, and lighting + post-processing. For 3D assets, don't settle for low-poly placeholders — generate and refine real models with Blender running headless (bpy scripts: build/import geometry, UV-unwrap, apply PBR materials, export glTF for Three.js). Generate textures with GPT Image for albedo/normal/roughness maps (chatgpt-imagegen "<prompt>" -o <path>.png --size 1024x1024 --timeout 300, one at a time).

3. QA — deploy live to Vercel, then use the Playwright CLI to screenshot the deployed build and compare each shot against the real Counter-Strike 2 reference screenshots. Fix every difference. Target clean tactical realism — crisp daylight, sharp real-time shadows, ambient occlusion, PBR materials, subtle bloom — NOT neon or stylized. Stack: Three.js / WebGL, WASD + mouse, click to shoot, smooth 60fps.

DONE when: the clone is deployed live on Vercel and, side by side, the Playwright screenshots of your build look as close to the real Counter-Strike 2

The exact SimmonsBench test prompt every model received.

The builds

Fable 5 · Screenshot of the buildRuns in your browser

Written by

Pat Simmons. Pat tests AI models and tools on his YouTube channel.

You might also like.

Articles, builds, skills and tools on the same idea.
Browse all builds

AI FOR MORTALS.

Your daily AI edge that will make everyone wonder how you do it all.

Every weekday, the practical AI updates that matter, and exactly what to do with them.

The latest in AI, in plain English. Free, every weekday.