The author says they used the same prompt to have GPT-6.2 Astra and GPT-6.2 Sol generate single-file night racing games, and shared two videos. The original post does not disclose the full prompt or runnable files, so playability cannot be verified.
Tasks suited to this: Comparing how two models turn a short game brief into a single-file web prototype.
Tasks not suited to this: Drawing conclusions about the full prompt, code quality, stability, or actual playability.
Applicable model versions: GPT-6.2 Sol. The original post also mentions GPT-6.2 Astra.
Applicable client, agent, or API: Not specified.
Recommended reasoning level and parameters: Not specified.
The original post provides only a task summary, not the full prompt. The following is an original task rewritten from that summary and should not be treated as the author's original prompt:
Rewritten example: Create a single-file HTML night racing prototype that opens directly in a modern desktop browser… This is a necessary excerpt; read the original source for full context.
The only workflow that can be confirmed from the original post is that the author gave GPT-6.2 Astra and GPT-6.2 Sol the same prompt to generate a single-file night city racing game without manual code edits. The author shared videos, with the red car representing Astra and the yellow car representing Sol.
Original post time: 2026-11-18 00:04 UTC (2026-11-18 in Singapore time).
The author described these requirements: the same prompt, 3D night city racing, a single HTML file, and no manual code edits.
The page shows two embedded videos, distinguishing the two models' results with a red car and a yellow car. It does not provide source code, downloadable files, or control instructions.
The original post does not disclose the full prompt, model parameters, generation time, or quantifiable evaluation results.
The videos show only that the author presented generated results. They do not prove that the files run, that the interactions are complete, or that no fixes are needed. Differences in color and visuals are also insufficient to draw conclusions about model quality. Do not treat the rewritten example as the prompt from the original post.
To reproduce the comparison, submit exactly the same original brief to both models, save the original HTML files, and check in the same browser whether they open, support keyboard controls, provide collision feedback, and can be restarted. The original post does not provide the original prompt or files, so the author's test cannot be reproduced exactly.
GPT-6.2 Sol