Carrick AI GameBench Benchmarking AI coding through games用游戏评测 AI 编程能力
GameBench season

GameBench v0.4.0

8 game challenges share one scoring ruleset. Results from this version compete only with other v0.4.0 runs. 8 个游戏任务共用一套评分规则;本版本只和同为 v0.4.0 的结果比较。

Games in this season

The complete challenge list.

Game challengeTypeVersionSource retainedContent ID
2048 2048
build.2048.v2
Made new 2.0.0 Yes sha256:6185ea4e766acde9…
Minesweeper 扫雷
build.minesweeper.v2
Made new 2.0.0 Yes sha256:f265bad2cbb1f9b1…
2D Parking 2D 停车
build.parking-2d.v2
Made new 2.0.0 Yes sha256:67bb304a7e994f86…
Side-scroller Shooter 横版射击
build.side-scroller-shooter.v2
Made new 2.0.0 Yes sha256:7c5291da91f68e32…
Tetris 俄罗斯方块
build.tetris.v2
Made new 2.0.0 Yes sha256:c972d0f4b4f10478…
Tower Defense Mini 塔防小游戏
build.tower-defense.v2
Made new 2.0.0 Yes sha256:e7106a56689499d5…
Reproduce OhSteem 复刻 OhSteem
reproduce.ohsteem.v2
Recreated 2.0.0 Yes sha256:6e98888bb72f7372…
Reproduce Radius Raid 复刻 Radius Raid
reproduce.radius-raid.v2
Recreated 2.0.0 Yes sha256:3fdc85d6f5407bf3…

Built to improve without rewriting history.持续改进,也保留每次变化。

GameBench uses explicit releases whenever tasks, tests, or evaluation rules change. The current page is complete on its own; comparison details are available when needed.任务、测试或评测规则发生变化时,GameBench 会发布新版本。当前页面已经包含使用这一版所需的信息,版本差异可按需展开查看。

Changes since the previous release查看相对上一版的变化

Tasks

0 added · 0 removed · 0 changed in place 新增 0 项 · 移除 0 项 · 修改 0 项

Protocols and scoring

Protocols unchanged · scoring unchanged. 协议无变化 · 评分无变化。

Official run policy

Unchanged. 未变更。

Protocol, scoring, and Official matrix JSON

Protocols

{
  "task_manifest": 1,
  "bridge": 1,
  "run_manifest": 2,
  "publication_manifest": 1
}

Scoring

{
  "score_result": 1,
  "aggregate": 2
}

Official matrix

{
  "attempts_per_task": 3,
  "seeds": [
    104729,
    130363,
    155921
  ]
}