Carrick AI GameBench Benchmarking AI coding through games用游戏评测 AI 编程能力
Choose your game

Eight challenges. Dozens of model-made versions.

Open a challenge to play different models solving the same brief. Six games are made from scratch; two ask the model to recreate an existing open-source game.

v0.4.0 · current

The test set evolves over time. Games shown here always come from the current release; earlier contracts remain available in the version archive. 测试集会持续迭代。这里默认只展示当前版本;较早的任务规则仍保留在版本档案中,方便审计。

About versioning了解版本迭代
Current season · v0.4.0

Pick a challenge.

Each card shows how many model-made builds are ready to play.

01

2048 2048

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 1 0 playable builds
02

Minesweeper 扫雷

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 1 0 playable builds
03

2D Parking 2D 停车

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 2 0 playable builds
04

Side-scroller Shooter 横版射击

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 3 0 playable builds
05

Tetris 俄罗斯方块

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 2 0 playable builds
06

Tower Defense Mini 塔防小游戏

Build a complete game from a precise playable brief. 按明确的需求,从零做出一款完整可玩的游戏。

make new difficulty 3 0 playable builds
07

Reproduce OhSteem 复刻 OhSteem

Recreate the behavior and look of an existing open game. 参考现有开源游戏,还原它的玩法和外观。

recreate difficulty 2 0 playable builds