One prompt, one shot per model, judged by spec fidelity, collision correctness, polish, and mobile/touch support. Part of our ongoing /design benchmark series in Command Code.
The prompt
Create a vertical Galaxy Shooter game as a single HTML file using Canvas. Portrait mobile style canvas (400x700). Player controls a cool blue futuristic spaceship at the bottom that moves left and right with arrow keys or on-screen touch buttons. The ship shoots lasers upward automatically or with tap/space. Enemies (colorful alien ships) come down from the top in formations. Destroy enemies and avoid their shots. Add glowing effects, particle explosions, and starfield background that scrolls down. Collect power-ups. Show score at the top, lives (hearts), and level. Include start screen and game over screen with restart. Make it fun, smooth, and visually vibrant like classic vertical shooters. Output only the full code.
Seen on X
An earlier three-model run of this exact prompt - Fable 5, Grok 4.5, and GPT 5.5 - was posted with the actual clip:
Fable 5 vs Grok 4.5 vs GPT 5.5 We put three top-tier models to build a same game challenge....
That real-play impression - Grok 4.5 reading as "polished," Fable 5 and GPT 5.5 as "rushed" - lines up with the code-read scorecard below, where Fable 5 and GPT-5.5 both score lowest on spec fidelity and polish.
Scorecard
| Model | Spec fidelity | Collision | Polish | Mobile/touch | Mean | Cost | Quality/$ |
|---|---|---|---|---|---|---|---|
| Kimi K3 | 5 | 5 | 5 | 5 | 5.00 | $0.030 | ~167 |
| Grok 4.5 | 5 | 4.5 | 5 | 4.5 | 4.75 | $0.09 | ~53 |
| Opus 5 | 5 | 5 | 4.5 | 4.5 | 4.75 | $0.24 | ~20 |
| Fable 5 | 3 | 5 | 4 | 5 | 4.25 | $0.38 | ~11 |
| GPT-5.5 | 4 | 3.5 | 2 | 3.5 | 3.25 | $0.11 | ~30 |
Method
n=1 per model, no re-rolls. Only Grok's build was actually hand-played; Fable 5's, GPT-5.5's, Kimi K3's, and Opus 5's rows come from a code read plus a headless smoke test, not from playing them.
Which model should you reach for
If you're building a mobile-style shooter and want the best result per dollar spent, Kimi K3 is the one to try first - a perfect score in this round at the lowest price in the table. Opus 5 is worth a look if you specifically need on-theme art direction and weapon variety, though check that a "best score" actually gets saved before you ship it; this build doesn't.
Small gaps like that are easy to miss until you actually read or play the output, which is the case for testing a couple of models on your own brief rather than picking one from a leaderboard. /design in Command Code makes that a /model swap away.
Verdict
Kimi K3 wins outright on both quality and price in this round - a 5.00 mean at $0.030 puts its quality-per-dollar well ahead of everything else in the table. Grok 4.5 and Opus 5 tie for second on raw quality (4.75), with Grok the cheaper of the two by a wide margin. Fable 5 is solid on collision and mobile support but loses points on spec fidelity. GPT-5.5 trails clearly across the board, with polish as its weakest axis.
If you're optimizing for "best game per dollar," Kimi K3 is the one to reach for first in this comparison.
Try this yourself
1npm i -g command-code1/model kimi-k3
2/design Create a vertical Galaxy Shooter game as a single HTML file using Canvas. Portrait mobile style canvas (400x700). Player controls a cool blue futuristic spaceship at the bottom that moves left and right with arrow keys or on-screen touch buttons. The ship shoots lasers upward automatically or with tap/space. Enemies (colorful alien ships) come down from the top in formations. Destroy enemies and avoid their shots. Add glowing effects, particle explosions, and starfield background that scrolls down. Collect power-ups. Show score at the top, lives (hearts), and level. Include start screen and game over screen with restart. Make it fun, smooth, and visually vibrant like classic vertical shooters. Output only the full code.Full command reference: commandcode.ai/docs/slash-commands/design. More one-shot comparisons: CommandCodeAI/slash-design-showcase.
