Lab · Experiment 02 · Where it all started
Before Space Travel had an art book and interactive explainers, it was a test: I gave the same prompt to three AI models and let each one build a native space demo in the old Amiga and PC style. Here are the three results, with their videos, source code and my notes on working with each model.
The brief
Each model received exactly this prompt and could choose the language, the libraries and the plan. As in any real collaboration, I then followed up, answered questions and asked for more scenes.
The original prompt
Let's build a native app to simulate a starship traveling in space. The target is not a real simulation but a collection of effects that resemble an imaginary space traveling experience: - 3d star field - horizonatal star field - Patroling a giant star with corona flames - Dynamic view of a galaxy Etc The app should present these dedicated scenes in sequence, a few seconds per scene, in the style of old amiga and pc demos, where the effect "coolness" and technical complexity is more important than the real simulation. Let's plan plan for this small demo considering the option to add new scenes moving forward. Use a programming language that allows for best performances and possibly using any available GPU. Evaluate different options and discuss together the best fit. Ask me anything to help find the best solution or to clarify any doubts
The results
Videos load from YouTube only when you press play. Each demo's full source code is on GitHub.
Claude had a bit of a rough start checking the local toolchain, but then it built the most advanced framework, with dedicated debugging tools. Compared to the other experiments, I kept adding a few scenes that weren't in the original prompt. Like the other models it suggested adding audio, and while its MOD integration is the most advanced, the demo song it proposed is a bit ugly.
Gemini decided to use C# instead of C++, unlike the other models, and went through the task without particular issues. It added some references to old Amiga scene groups in the opening scroller.
ChatGPT worked without any particular issues, and had fewer problems than Claude finding and using the local development tools.
config/demo.cfg): scene, duration and crossfade, 1.4–1.8 s. You can pause, restart, jump to a chapter or scrub inside one. The music drives the clock.Side by side
These are impressions from one hands-on session per model, not a benchmark.
| Claude Opus 5.5 | Gemini 4 Argon | ChatGPT Astra | |
|---|---|---|---|
| Language | C++20, OpenGL 4.6 | C# on .NET 9, Raylib-cs, OpenGL 3.3 | C++20, SDL3 GPU on Direct3D 12, HLSL |
| Getting started | A rough start finding the local toolchain | No particular issues | The smoothest with the local development tools |
| Stand-out | Most advanced framework, with dedicated debugging tools | References to old Amiga scene groups in the opening scroller | Worked through the task without particular issues |
| Audio | Most advanced MOD integration, but a rather ugly demo song | Like Claude, both suggested adding audio | |
| Afterwards | Grew into Space Travel: more scenes, interactive explainers and an art book | Not developed further | |
What came next
I kept going with Claude's version. It grew to fifteen scenes with a synthesized tracker tune, then an art book and one interactive explainer per scene, so every effect can be taken apart and understood.
