Output Loop Bench · Artifacts
Game: AI models compared
A complete playable game in one file. 19 models tested on the same task.
Pipe puzzle game (5 levels)
A complete game in one file tests the game loop, input, state and difficulty in one go. It is easy to judge: either level 5 is playable or it isn't.
| Model | Maker | Time | Tokens |
|---|---|---|---|
| claude-sonnet-5-5 ★ Output Loop's pick | Anthropic | 369.2s | 138,279 |
| nemotron-3-nano-4b | NVIDIA | 23.1s | 2,542 |
| composer-2.5 | Cursor | 40.9s | 28,928 |
| muse-spark-1.3-medium | Meta | 45.1s | 31,429 |
| gemini-4-argon-medium | 88.4s | 28,736 | |
| gpt-6-astra | OpenAI | 137.1s | 21,648 |
| kimi-k3-max | Moonshot (Kimi) | 152.7s | 48,166 |
| gpt-6.1-sol | OpenAI | 165.4s | 21,977 |
| MiniMax-M2.5 | MiniMax | 178.4s | 45,920 |
| gemini-3.1-pro | 201.7s | 24,716 | |
| bonsai-27b | Prism ML | 223.3s | 12,195 |
| gemma-4-26b-a4b-qat | 232.7s | 25,984 | |
| deepseek-v4-pro | DeepSeek | 352.1s | 65,527 |
| kimi-k3 | Moonshot (Kimi) | 396.1s | 23,447 |
| claude-fable-5-1 | Anthropic | 402.4s | 124,234 |
| claude-opus-5-5 | Anthropic | 433.2s | 137,487 |
| muse-glimmer | Meta | 648.5s | 19,754 |
| grok-4.7 | xAI | 1021.2s | 87,449 |
| glm-5.3 | Z.ai | 1148.0s | 60,409 |