WebsiteBenchWebsiteBench

LegalNature

Rebuilt by GPT 6.1 Sol: 144 of 389 hidden tests passed (37.0%)

reference.local/
Reference for LegalNature, HomeReference
rebuild.local/
Rebuild by GPT 6.1 Sol, HomeGPT 6.1 Sol
    Every model on LegalNature
    ModelPassed%Screenshot
    GPT 6.1 Sol14437.0
    DeepSeek V4.1 Flash13935.7
    GLM 5.3 Flash13534.7
    Claude Opus 5.513434.4
    Claude Opus 5328.2
    Kimi K3246.2
    Qwen 3.8 Max235.9
    Grok 4.6112.8

    Every result in the WaffleRun the benchmark