Leaderboard

The standard suite: 9 builds, the same prompt and settings for every model.

How models are ranked

Every model gets the same 9 tasks at its default effort, builds blind (it never sees a render of its result) and may fix up to 2 crashes. Models are ordered by how many tasks produced a working asset, then by how many did so without a crash, then by cost. Nothing here scores how the results look: that is yours to judge in the scene below and in the arena. Cost and time are totals over the whole suite, failed runs included. This is suite version 1.

Models ranked on the standard suite
RankModelSuite tasksBuiltWithout a crashSuite costSuite timeAll runsWorked
1stClaude Opus 5.5Barrel, voxel: built first tryBarrel, N64: built first tryBarrel, realistic: built first tryChest mimic: built first tryDragon whelp: built first tryRobot: built first tryAnime girl: built first tryPirate ship: built first tryGraveyard: built first try9 of 99$1.8114 min4398%
2ndGemini 4 CarbonBarrel, voxel: built first tryBarrel, N64: built first tryBarrel, realistic: built first tryChest mimic: built first tryDragon whelp: built first tryRobot: built first tryAnime girl: built first tryPirate ship: built first tryGraveyard: built first try9 of 99$2.6158 min37100%
3rdGPT-6 AstraBarrel, voxel: built first tryBarrel, N64: built first tryBarrel, realistic: built first tryChest mimic: built first tryDragon whelp: built first tryRobot: built first tryAnime girl: built first tryPirate ship: built first tryGraveyard: built first try9 of 99$3.2519 min44100%
4thClaude Fable 5.2Barrel, voxel: built first tryBarrel, N64: built first tryBarrel, realistic: built first tryChest mimic: built first tryDragon whelp: built first tryRobot: built first tryAnime girl: failedPirate ship: built first tryGraveyard: built first try8 of 98$17.7974 min4693%
5thGPT-6.2 AstraBarrel, voxel: built first tryBarrel, N64: built first tryBarrel, realistic: built first tryChest mimic: not runDragon whelp: not runRobot: not runAnime girl: not runPirate ship: not runGraveyard: not run3 of 93$0.272 min12100%
Claude Sonnet 5.5Not run yet
Claude Haiku 5.5Not run yet
GPT-6.1 SolNot run yet
GPT-6.1 TerraNot run yet
GPT-6 LunaNot run yet
Gemini 4 NeonNot run yet9100%
Gemini 3.8 FlashNot run yet
Grok 4.7Not run yet
DeepSeek V4.1 FlashNot run yet
Qwen 3.8 Max PrimeNot run yet
Kimi K3.1 ProNot run yet

The whole suite in one scene

View

39 of 39 runs shown

Model
Task
Play on every model
Scene

Every run is on the floor at once. Drag to orbit; pinch, or hold Ctrl or Cmd and scroll, to zoom. Click a model to inspect it. All models are scaled to the same height.