CoffeeBench: Benchmarking Long-Horizon LLM Agents in Heterogeneous Multi-Agent Economies

  • 类型:arxiv
  • 标识:2606.16613
  • 链接:https://arxiv.org/abs/2606.16613
  • 主分类:agent
  • 形态:benchmark
  • 被引:6
  • 被引来源:Semantic Scholar
  • S2被引:6
  • OpenAlex被引:0
  • 影响力被引:0
  • TLDR:Analysis of agent behavior reveals substantial differences in long-horizon economic interaction: higher-performing models communicate more actively with other firms, whereas Claude~Haiku~4.5 exhibits an idle-drift failure mode, repeatedly choosing inaction despite producing coherent assessments and plans.
  • OpenAlex ID:W7164882062
  • OpenAlex DOI:10.48550/arxiv.2606.16613
  • DOI:10.48550/arxiv.2606.16613
  • DOI来源:OpenAlex
  • 开放获取:green
  • 开放获取链接:https://doi.org/10.48550/arxiv.2606.16613
  • OpenAlex更新:2026-07-19
  • 副分类:evaluation
  • 待LLM分类:否
  • 标题中文:CoffeeBench:面向异构多 Agent 经济中的长视野 LLM Agent 基准测试
  • TLDR中文:对 Agent 行为的分析揭示了长视野经济交互中的显著差异:表现更好的模型与其他企业的沟通更为活跃,而 Claude Haiku 4.5 则表现出 idle-drift 失效模式,在生成连贯评估与规划的同时仍反复选择不行动。
  • 来源文件
  • /inbox/tom/_candidates/2026-06-29-agent-memory-tool-use-candidates.json
  • /inbox/tom/_candidates/2026-06-28-agent-rag-longcontext-candidates.json
  • [S2 enrich]
  • [OpenAlex backfill]