WebsiteBenchWebsiteBench

Ollie

Rebuilt by Claude Opus 5: 30 of 394 hidden tests passed (7.6%)

reference.local/
Reference for Ollie, HomeReference
rebuild.local/
Rebuild by Claude Opus 5, HomeClaude Opus 5
    Every model on Ollie
    ModelPassed%Screenshot
    Claude Opus 5307.6
    Claude Opus 5.5307.6
    GPT 6.1 Sol287.1
    GLM 5.3 Flash256.3—
    Grok 4.6235.8
    Kimi K3164.1
    DeepSeek V4.1 Flash71.8
    Qwen 3.8 Max00.0—