Hugging Face、Open Agent Leaderboardを公開
Hugging FaceがOpen Agent Leaderboardをリリース。コーディングやカスタマーサポートを含む6つのタスクにわたってAIエージェントシステム全体を評価するベンチマーク。
hugging face launched the open agent leaderboard, a new open benchmark that evaluates complete ai agent systems, incorporating their tools, planning, and memory. unlike standard model benchmarks, it tests agents on six diverse tasks including real-world coding via swe-bench verified and complex web research via browsecomp+. the initiative uses the exgentic framework for reproducible evaluations and highlights the trade-offs between generality, performance, and cost.