Developer demos TypeSafe's Jev model generating platformer levels live
A blog experiment feeds game-state data to the newly launched Jev structured-output model to generate runner-game terrain on the fly.
What to know
- Jev, launched by TypeSafe on September 15, 2026, is a zero-shot classifier model optimized for fast, cheap structured decisions rather than free-text generation.
- A developer used it to generate platformer terrain in real time for a neon-night runner-game prototype, feeding it game-state snapshots and multiple-choice terrain options in a single API call.
- Latency is sub-second but not sub-100ms, and only input tokens carry a cost ($0.042 per million); the author called it promising but still exploratory for game development.
TypeSafe Creator of the Jev modelHugo Duprez Developer / blog author (Sprite Fusion)
How it unfolded 3 developments, newest first · click a bar or a number to jump posts
-
3
Story resurfaces on Hacker News under a new submission
The same write-up is resubmitted to Hacker News by a different user, giving the experiment a second round of visibility.
-
2
Duprez publishes the Jev level-generation write-up
The Sprite Fusion blog post detailing the experiment—sending game-state snapshots to Jev to choose terrain widths, gaps, heights and surface types—is submitted to Hacker News as a Show HN.
“Level generation keeps working fine.”
— HugoDz (blog author) -
1
Duprez shows Jev generating game levels in real time
Hugo Duprez posts a demo on X showing Jev filling in runner-game terrain live, framing cheaper structured output as significant for game development.
“Jev can generate game levels in real time. Faster and cheaper structured output could be a big deal for game dev!”
— @HugoDuprez -
Jev can generate game levels in real time. Faster and cheaper structured output could be a big deal for game dev!
-
-
background
TypeSafe launches the Jev structured-output model — TypeSafe introduces Jev, a zero-shot classifier model that returns structured decisions with probabilities instead of raw text, priced at $0.042 per million input tokens with free output tokens.