Public benchmark

AgentArk Bench on Kaggle

AgentArk Hub no longer maintains a separate leaderboard. Kaggle Bench is the canonical public result surface.

14benchmark tasks9evaluated models

#1GPT-6 Astra95%

#2Gemini 4 Argon94%

#3GPT-5.6 Sol79%

#4Gemini 3.8 Flash72%

#5GPT-5.568%

#6Claude Opus 568%

#7Gemini 3.5 Flash49%

#8Claude Opus 4.843%

#9Grok 4.20 Reasoning32%

Snapshot refreshed Oct 24, 2026, 11:36 AM. Follow the Kaggle link for live task-level results and benchmark details.