The Heptagon Bouncing Balls Test

Pick a version

Seven models, one physics prompt. Whose balls stay inside?

• Qwen 3.8 MaxCurrentTime 15m 9sAPI cost $0.24
• Claude Fable 5.1Time —API cost —
• Gemini 4 ArgonTime 46sAPI cost $0.044
• Claude Opus 5.5Time —API cost —
• Kimi K3Time 6m 5sAPI cost $0.32
• GPT-6.1 SolTime 3m 28sAPI cost —
• GPT-6 AstraTime 4m 10sAPI cost —

From the video10 AIs Play Battle Royale