Raúl Berná · 3 min read
Same prompt, two models: Fable 5.2 vs Qwen4-Max
I gave Qwen4-Max the same brief I gave Fable 5.2 for Duskline. Its game is worse, but it cost $0.67 instead of about $16. That makes me think about when paying more is worth it.
Last week I asked Claude Fable 5.2 for a 3D game in a single HTML file: a robot escaping a drone along an elevated metro line at sunset. That became Duskline. This weekend Alibaba opened Qwen4-Max, its new large model, to everyone, and today I gave it the exact same prompt.
Its game is called Metro Dash, and you can play it here: metro-dash-qwen.vercel.app.
How I compared them#
Both worked with the same tools: Claude Code with permission to read and write files, and nothing else. No browser, no shell and no internet. Neither could open its game while making it: they wrote it, reread it and handed it over.
The only difference is that Fable ran with high effort and Qwen used its default reasoning.
| Fable 5.2 (Duskline) | Qwen4-Max (Metro Dash) | |
|---|---|---|
| Time | 45 min | 19 min |
| Cost | about $16 at API prices | $0.67 |
| File size | 93 KB | 47 KB |
| Errors when opened | none | none |
Fable's $16 is what it would have cost through the API; I paid for it with my subscription. Qwen's $0.67 is what OpenRouter charged.
I played it#
Metro Dash is worse. The robot and the tracks have much less detail, the game over screen only shows the score, and overall it feels less finished. Duskline has sleepers, overhead wires, a robot you can recognise and a final screen with distance and time.
But it works first time. There are parked trains and oncoming trains, sparks, a shield, a warning when the drone gets close, and a saved best score. For a game nobody opened before handing it to me, that's a lot.
What I take from it#
I don't think this says which model is better. It's one game and one attempt per model, and a different prompt could flip the result.
What does make me think is the price. If I want a prototype to see whether an idea works, $0.67 is plenty, and I can ask for ten versions for what one from Fable costs. If I want something I'm going to show people, I'd rather pay more and get it right the first time.
And one thing still surprises me: both wrote a WebGL game by hand, with no libraries and no way to test it, and both worked.
The code for Metro Dash and the prompt are on GitHub. If you play both, I'd like to know which one you think is better.


