做 AI 智能体评估或选型的团队终于有了基于真实使用数据的排行榜，比传统基准测试更贴近实际效果，值得关注。
LMSYS 推出 Agent Arena，一个基于真实用户交互的智能体能力排行榜。该排行榜通过因果追踪方法，分析智能体在竞争情报、市场分析、科研等深度研究任务中的表现。排行榜依据五个行为信号（确认成功、表扬/投诉、可操控性、故障恢复、工具幻觉）动态更新。用户每次使用 Agent Mode 的会话都会影响排名，使评估更贴近实际使用场景。
ICYMI: Agentic AI is now measured in the Arena. Agent Mode can handle deep research around competiti...
ICYMI: Agentic AI is now measured in the Arena. Agent Mode can handle deep research around competitive intelligence, market sizing & opportunity analysis, scientific & medical research and more. Every session shapes the Agent Arena leaderboard. Get a walkthrough of the causal tracing methodology with Evan. Dive into the thread for more on Agent Mode and Agent Arena. 0:00 How causal tracing works 1:09 A living leaderboard that evolves with AI 1:35 The five behavioral signals explained 1:54 Confirmed success 2:22 Praise and complaint 2:46 Steerability 3:13 Bash recovery 3:39 Tool hallucination 4:11 Natural language model insights 4:37 Per-signal leaderboard cards walkthrough 5:41 What people actually do in Agent Arena 6:01 Scale: conversations, tool calls, and context length 6:13 Most-used tools and task types 7:22 Why this is a real-usage leaderboard 7:49 Labs comparison: OpenAI vs. Anthropic vs. the field 8:24 How Agent Arena differs from past evaluations Your browser does not support the video tag. 🔗 View on Twitter 💬 1 🔄 2 ❤️ 15 👀 1454 📊 3 ⚡