WebsiteBenchWebsiteBench

LinkedIn

Rebuilt by DeepSeek V4.1 Flash: 140 of 565 hidden tests passed (24.8%)

reference.local/
Reference for LinkedIn, HomeReference
rebuild.local/
Rebuild by DeepSeek V4.1 Flash, HomeDeepSeek V4.1 Flash
    Every model on LinkedIn
    ModelPassed%Screenshot
    DeepSeek V4.1 Flash14024.8
    Claude Opus 513023.0
    GPT 6.1 Sol12622.3
    GLM 5.3 Flash12522.1—
    Kimi K310318.2
    Claude Opus 5.510218.1
    Qwen 3.8 Max7813.8
    Grok 4.67613.5

    Every result in the WaffleRun the benchmark