Breaking

Same Prompt Tested Across 13 Claude Models Over 14 Months to Build a Procedural 3D Terrain Game

November 4, 2026 at 21:15 EST

Software Dev & Coding
Foundation Models

A developer has reported re-running one identical one-shot prompt, asking for an infinite procedural 3D terrain game, against all 13 Claude models released over a 14-month span, with the earliest and latest outputs showing how quickly AI-generated code has improved.

14-month span · Claude model series

One unchanged prompt, 13 Claude models, and a visible leap in one-shot code generation

A developer ran the same request for an infinite procedural 3D terrain game against every Claude model in a 14-month window, then compared the earliest and latest results.

Claude models tested with the identical prompt

~14

Months between the earliest and latest model

Prompt, unchanged across every run, with no follow-up iteration

Coverage of the test: 13 models across the window

Each column is one model in the series, shown in release order. Columns are equal height because the test measures count, not a score.

Earliest model Latest model

Earliest output

First model in the series

The starting point of the comparison, using the same unchanged prompt.

Latest output

Most recent model in the series

The developer's assessment is that this output shows far stronger one-shot results.

How the test works

Fixed prompt
Same text every run

→

Claude model
Each of 13 in turn

→

3D terrain game
One-shot code output

→

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.