ProphetLab

PROPHETLAB / BENCHMARKS

真实的任务与环境，真正的评测与检验。Real tasks and environments.Real benchmarking and evaluation.

用真实任务检验模型如何与环境交互，寻找证据，综合分析，并交付结果。这里汇集 ProphetLab 已公开发布的基准。Real tasks test how models interact with their environment, find evidence, bring it together through analysis, and deliver results. Explore ProphetLab’s published benchmarks.

已发布基准PUBLISHED BENCHMARKS01
已发布PublishedFINANCE / V1

Investor's
Last Exam v1

面向智能体的定点时刻股票研究基准。在相同的数据与工具环境中，观察从信息检索到研究交付的完整过程。A point-in-time equity research benchmark for agents, examining the full path from retrieval to research deliverables in a shared data and tool environment.

  • 信息截止时间由环境约束Information cutoffs enforced by the environment
  • 明确评分标准与可复核轨迹Explicit criteria and inspectable trajectories
  • 公开展示结果与方法Public results and methodology
查看基准与结果Explore the benchmark方法与任务构成Tasks & methodology阅读研究文章Read the research story

当前公开目录包含 Investor's Last Exam v1。任务数量、模型版本和评分口径以基准详情页的对应发布快照为准。This public catalogue currently includes Investor's Last Exam v1. Refer to its release snapshot for task counts, model versions, and scoring methodology.

有新的评测需求？A different evaluation question?

欢迎模型团队与领域专家就数据、任务设计及评测合作与我们沟通。Get in touch about data, task design, and evaluation partnerships. We welcome model teams and domain experts.

洽谈合作Contact us