11 Mind‑Blowing GPT‑5.6 Design Cases That Showcase Its New Capabilities

The author was excited that GPT‑5.6, previously limited to subscribers, is now fully open and merged with Codex, eliminating the need to switch between tools.

1. Voxel‑Style Manhattan

Source @阿良 used GPT‑5.6 Sol to generate a voxel‑style Manhattan entirely autonomously over nearly a week. The result’s detail was described as "outrageously high."

2. Controlling Blender

Japanese creator @AI様の下僕 prompted GPT‑5.6 Sol to operate Blender, producing a series of rendered images.

3. City Islands

@thehype tested three prompts with a single‑file HTML using three.js to create floating island districts for London, Paris, and New York. Requirements included square islands, steep cliffs, waterfalls flowing from drainage pipes, first‑person flight camera, dense landmark distribution, cinematic golden‑hour sky, glow effects, and no UI or text.

Results (combined for three cities):

Cost: grok 4.5 – $0.41; glm 5.2 – $1.45; gpt 5.6 sol pro – $4.12; fable 5 – $11.37

Tokens: grok 4.5 – 60,291; fable 5 – 174,083; gpt 5.6 sol pro – 241,739; glm 5.2 – 243,188

Code lines: fable 5 – 2,830; gpt 5.6 sol pro – 3,003; glm 5.2 – 3,474; grok 4.5 – 3,830

4. 3D Globe Dashboard

@Kai observed that GPT‑5.6 Sol represents a major leap compared with 5.5, while Fable 5 still excels in spacing, balance, and first‑draft polishing. Sol achieves comparable quality at roughly half the cost.

5. Procedural Terrain Generator

@Majid Manzarpour built a procedural terrain generator using only prompts, without invoking any agent skills.

6. Google‑Earth Clone

@Pankaj Kumar prompted GPT‑5.6 Sol Ultra to construct a Google‑Earth clone featuring 3D terrain, satellite imagery, cities, landmarks, real‑time weather, day‑night cycle, cinematic navigation, and global search. The project consumed over 30 million tokens, which the author deemed worthwhile.

7. UI Replication

@The Bugged Dev attempted a UI replica but felt the result did not meet expectations.

8. Kyoto Street Scene

@gimu asked GPT‑5.6 to generate a background image of a Kyoto street; the resulting cobblestones and lighting texture were praised.

9. 3D Watch

@Chetaslua spent 20–40 minutes creating a 3D watch model.

10. GTA‑6 Prototype

@チャエン (CEO of デジライズ) gave GPT‑5.6 Sol Ultra the prompt "make a GTA‑6" and received a result with noticeably better visuals and design than GPT‑5.5.

11. Xiaohongshu Clone

@噪点noisepoint recreated a Xiaohongshu‑like app in a few minutes, including search, like, and collect features, and deployed it as a website.

Overall, the eleven cases demonstrate that GPT‑5.6’s design capabilities have taken a significant step forward, achieving high completion quality. The author advises planning projects before prompting to avoid exhausting token quotas, and invites readers to share their own GPT‑5.6 creations for collective learning.