Jan gave Gemini 4 Neon and GPT-6.2 Sol the same one-line English prompt to build a vampire cathedral, then showed both buildings inside and out in a side-by-side video.
Suitable tasks. Use a one-line prompt to prototype a cathedral in a scene-building tool and compare how models handle the building's silhouette and interior.
Unsuitable tasks. Deliverables that require a specific game, export format, dimensions, or gameplay mechanics.
Applicable model versions. Gemini 4 Neon and GPT-6.2 Sol, as labeled by the author.
Applicable clients, agents, or APIs. Unspecified.
Recommended reasoning level and parameters. Unspecified.
Build a huge haunted vampire cathedralThe author says both models received the same prompt. The original post shows one result per model. It gives no details on the number of generations, tool permissions, or export method.
The 42-second comparison video opens with exterior views of the block-style cathedrals, then moves inside. Both have spires, stained-glass windows, and carpeted naves. Neon has red hangings in the nave. Sol has a dark red roof and a purple pattern on the nave carpet.
| Position in the video | Model as labeled by the author | Cost reported by the author |
|---|---|---|
| Left | Gemini 4 Neon | $2.67 |
| Right | GPT-6.2 Sol | $1.69 |
The author named no winner and provided no billing breakdown.
The prompt specifies only scale, atmosphere, and the type of building. The video shows the exteriors and interior layouts. The post includes no runnable scene files or tool-call logs, and the camera movement leaves it unclear whether users can control the scene. The reported costs apply only to this comparison.
We used the international edition of Tabbit to view the original English post and video when collecting this example. We did not rerun either model. To repeat the comparison, submit the original prompt to each model using the same scene-building tool, starting project, and tool permissions. Record each model's reasoning level, number of generations, and actual cost.
Gemini 4 Neon