Fable 5 vs GPT 5.5 vs GLM 5.2 vs DeepSeek v4 Pro All these examples were one-shotted, I used the same prompt for all of them....
Four models, one prompt, one shot each with /design in Command Code.
The prompt
Create a cute retro pixel art 2D platformer game as a single HTML file using Canvas. Wide horizontal canvas. Player is a small blue character that can run left/right with arrow keys, jump with space. Include a pink cat companion that follows the player. Platforms to jump on, collect glowing blue diamonds for points, avoid falling off. Add a blue door as level goal. Simple gravity and smooth jumping physics. Display score, lives (hearts), level info at the top. Background with light purple sky and clouds. Text "RUN JUMP AND STOMP!" at start. Game over and win screen. Make it fun, colorful, and 8-bit style. Output only the full code.
What we found
Fable 5 vs GPT 5.5 vs GLM 5.2 vs DeepSeek V4 Pro
All these examples were one-shotted, I used the same prompt for all of them.
For the OS model, I used /design from Command Code.
All the designs look similar, but there's a huge difference in UX and pricing.
Cost breakdown:
The framing here is the useful part: on a fully-specified prompt like this one - specific character, specific companion, specific colors, specific win condition - visual output converges hard across models. Four different labs landed on results close enough to look "similar" at a glance. The differentiation moved to two places a screenshot doesn't show: how the game actually plays (UX), and what it cost to generate. That's consistent with the pattern this series keeps finding - the more prescriptive the prompt, the less room there is for visual divergence, and the more the real comparison is about feel and price rather than looks.
Try this yourself
1npm i -g command-code1/model fable-5
2/design Create a cute retro pixel art 2D platformer game as a single HTML file using Canvas. Wide horizontal canvas. Player is a small character that can run left/right and jump. Include a companion animal that follows the player. Platforms to jump on, collect glowing diamonds for points, avoid falling off. Simple gravity and smooth jumping physics. Score, lives, and level info at the top. Game over and win screens. Fun, colorful, 8-bit style.Swap /model to compare against GPT 5.5, GLM 5.2, or DeepSeek V4 Pro. Full command reference: commandcode.ai/docs/slash-commands/design. More one-shot comparisons: CommandCodeAI/slash-design-showcase.
