iCreatAI News
Skip to content
Trending storyWatching

Kradle releases Minecraft multi-agent evaluation benchmark

2 reports1 sources2 days agoUpdated

Understand this story

AI overview

Kradle.ai released a multi-agent AI evaluation benchmark, KradleBench, which tests models in Minecraft scenarios across five aspects: collaboration, economy, conflict, logic, and creativity. On the overall leaderboard, OpenAI GPT-6.2 Astra ranks first with 78.28/100, at a per-run LLM cost of $1.26; Google Gemini 4 Hydrogen and Anthropic Claude Opus 5.5 rank 7th and 8th respectively. Kradle says most AI benchmarks are static, while the real world is not.

AI overview based on source reports · Updated 2 days ago

Coverage timeline

Follow the coverage to explore different perspectives.

Nov 19
  1. Hacker News AI
    Show HN: KradleBench – Multi-agent AI evals in Minecraft

    KradleBench is a Minecraft-based benchmark that evaluates multiple AI models on collaboration, economy, conflict, logic and creativity tasks, with every run playable back as an interactive replay. Skilled human player SethBling beats frontier models at economy and conflict but trails top models on solo logic and creativity challenges, and GPT-6.2 Astra ranks first overall at 78.28/100 with $1.26 LLM cost per run.

  2. Hacker News AI
    KradleBench: Multi-agent AI evals in Minecraft

    KradleBench evaluates top AI models on multi-agent tasks in Minecraft, finding that models coordinate well and generate recognizable objects but that a skilled human player, SethBling, still beats them at strategy and conflict. GPT-6.2 Astra ranks first overall at 78.28/100 with a $1.26 LLM cost per run, ahead of Gemini 4 Hydrogen (43.90) and Claude Opus 5.5 (39.12). Each run can be replayed interactively, and one highlighted run shows GPT-6.1 Sol turning its diamonds into a sword and armor to eliminate every rival.

Activity for this story

Current comparable activity
3
Peak comparable activity
10Nov 19 02:00
Change in the past 24 hours
-49%

Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.