Hacker News · AI·· 1 d agoAI score55
KradleBench benchmarks multi-agent AI models in Minecraft
KradleBench: Multi-agent AI evals in Minecraft
AI summary
KradleBench, a benchmark from Kradle, evaluates frontier models on multi-agent tasks in Minecraft across collaboration, economy, conflict, logic and creativity. GPT-6.2 Astra ranks first overall at 78.28/100 with an LLM cost of $1.26 per run, while skilled player SethBling still beats the models at economy and conflict but trails them on solo logic and creativity challenges. Every run can be replayed as an interactive replay.
Source:Hacker News · AI · kradle.ai