Grok 4.5 vs Fable 5 vs GPT 5.5 vs GLM 5.2 vs GPT-5.6 Sol vs Kimi K3: guess the model

The same prompt, run twice, across two overlapping sets of models - six in total. Part of our ongoing /design benchmark series in Command Code.

The prompt

Create a retro pixel art horizontal space shooter game as a single HTML file using Canvas. Vertical or wide canvas, small orange spaceship that can move up, down, left, right. Smooth movement with slight inertia. Shoot with spacebar, collect glowing blue gems for points, avoid obstacles. Starry black background with scrolling cave walls (top and bottom). Simple enemies or obstacles, score display, lives, game over screen. Classic 8-bit/16-bit pixel style with nice glow effects on gems and ship thrusters. Output only the full code.

Round one: guess the model

Create these games in one shot, guess the model?

Grok 4.5, Fable 5, GPT 5.5, GLM 5.2

No scores attached to this round - it was posted as a blind challenge: four models, four outputs, guess which is which. That's a genuinely useful format for this kind of content: it forces the reader to actually look at what each model produced instead of skimming a ranking, and it's an honest test of whether a model's "style" is recognizable at all on a fixed brief.

Round two: four models, real findings

NR

Naymur Rahman

@naymur_dev

Fable 5 vs GPT-5.6 Sol vs Grok 4.5 vs Kimi K3 I ran the same /design prompt in Command Code across 4 model...

6:01 PM·Jul 21, 2026

This round swapped GPT 5.5 and GLM 5.2 for GPT-5.6 Sol and Kimi K3, and this time came with a real finding: Kimi K3 was the only one of the four to build out actual progression - leveling up, escalating bullet patterns, and a boss fight at the end of each level - on a prompt that only asked for "simple enemies or obstacles." That's the same pattern this series has seen from Kimi K3 elsewhere (the pixel-art space shooter benchmark in this series credits Grok 4.5 with the deepest progression system on a closely related prompt) - models given room in the brief will sometimes just build more game than was asked for, and it's worth checking for that rather than assuming every one-shot output covers the same scope.

Six models, one prompt

Across both rounds: Grok 4.5, Fable 5, GPT 5.5, GLM 5.2, GPT-5.6 Sol, and Kimi K3 have all now built this exact brief. Grok 4.5 and Fable 5 appear in both rounds, giving a rough consistency check across two separate one-shot generations rather than a single sample per model.

Try this yourself

Full command reference: commandcode.ai/docs/slash-commands/design. More one-shot comparisons: CommandCodeAI/slash-design-showcase.