WebsiteBenchWebsiteBench

FreshDirect

Rebuilt by Claude Opus 5.5: 292 of 990 hidden tests passed (29.5%)

reference.local/
Reference for FreshDirect, HomeReference
rebuild.local/
Rebuild by Claude Opus 5.5, HomeClaude Opus 5.5
    Every model on FreshDirect
    ModelPassed%Screenshot
    Claude Opus 5.529229.5
    GPT 6.1 Sol24624.8
    GLM 5.3 Flash22122.3
    Claude Opus 518418.6
    Kimi K315315.5
    DeepSeek V4.1 Flash13313.4
    Qwen 3.8 Max13013.1
    Grok 4.612712.8