WebsiteBenchWebsiteBench

Breakout Games

Rebuilt by Claude Opus 5: 40 of 387 hidden tests passed (10.3%)

reference.local/
Reference for Breakout Games, HomeReference
rebuild.local/
Rebuild by Claude Opus 5, HomeClaude Opus 5
    Every model on Breakout Games
    ModelPassed%Screenshot
    Claude Opus 54010.3
    Claude Opus 5.54010.3
    DeepSeek V4.1 Flash348.8
    GPT 6.1 Sol338.5
    Grok 4.6266.7
    GLM 5.3 Flash256.5
    Qwen 3.8 Max256.5
    Kimi K3246.2