Blind: prompts used no knowledge of the challenge code or other players' solutions.Honor system. Every prompt can be reviewed, so anyone can check.

Inference time depends on the hardware used.

RankModelInferenceQuantPromptsPlayerHardwareRun
1qwen3:4b-instruct7.6 s4,983 tokensQ4_K_M1HumRobin1302Apple M4 Max, 48 GBView
2qwen3:4b-instruct9.9 s10,040 tokensQ4_K_M1PlayValley1232Apple M4 Max, 48 GBView
3qwen3:4b-instruct10.3 s8,913 tokensQ4_K_M1TangleKite2056Apple M4 Max, 48 GBView
4qwen3:4b-instruct10.5 s9,151 tokensQ4_K_M1ZoomLinen9140Apple M4 Max, 48 GBView
5qwen3:4b-instruct12.1 s7,361 tokensQ4_K_M1KeepShell1029Apple M4 Max, 48 GBView
6qwen3:4b-instruct12.8 s9,606 tokensQ4_K_M1WhiskWind9084Apple M4 Max, 48 GBView
7qwen3:4b-instruct17.7 s8,280 tokensQ4_K_M1Hugo CornellierApple M4 Max, 48 GBView
2026-09-23ZoomLinen914010.5 s · 9,151 tokens
2026-09-23HumRobin13027.6 s · 4,983 tokens (-3.0 s)