Investor's Last Exam · 86 tasks

Equity research tasks and the financial environment 股票研究题目与金融环境

Twenty-five task types across seven research and market-forecast families, instantiated as 86 tasks with access to financial data up to each task’s own cutoff. 七类投研与行情任务、二十五种题型，共 86 道题；每道题只使用其数据截止时间前可获取的金融信息。

Investor's Last Exam is ProphetLab’s dataset and research environment for financial forecasting and equity research. Tasks connect evidence, financial reasoning, scenario analysis and probabilistic forecasts. Investor's Last Exam 是 ProphetLab 面向金融预测与投资研究构建的数据集与研究环境，考察从证据检索、财务推理到情景分析与概率预测的综合能力。

01 / the task set题目集合

Seven families, twenty-five task types七类任务，二十五种题型

Seven task families, including integrated research and market forecasts. Tasks are checked for research depth, scoring coherence and discrimination in pilot results; 86 question briefs are public.七类投研与行情任务。题目按研究深度、评分一致性与试测区分度检查，公开展示86道研究与预测题面。

Expand a task type to read its public examples. Task counts cover all 86 tasks.展开题型可查看公开示例。题量按全部 86 题统计。

Research-informed task design面向投研需求的题目设计

ProphetLab’s AI post-training and investment research teams develop the tasks with input from investment practitioners on questions and scoring criteria. Each task has its own answer requirements, covering conclusions, calculations and supporting evidence. 题目由 ProphetLab 的 AI 后训练与金融投研团队建设，一线投资从业者参与问题与评分标准设计。每道题分别设置答案要求，考察结论、计算过程与支持证据。

Practitioners help select tasks that analysts can complete but remain challenging for frontier models, including portfolio adjustments under risk constraints, implied-expectations analysis and pre-earnings forecasts. 从业者参与筛选分析师能够完成、但前沿模型仍具挑战的任务，例如风险约束下的持仓调整、股价隐含预期计算与财报发布前的业绩预测。

Research across companies and periods跨公司、跨期间的长程研究

Tasks require multiple rounds of evidence gathering and validation across companies and reporting periods. Agents connect financial disclosures with supplier and customer relationships, supply and demand, and competitive conditions to revise their forecasts. 任务要求模型跨公司、跨财务期间开展多步取证与反复验证，将财务披露与上下游关系、供需变化和竞争格局联系起来，形成并修正预测。

For example, an external advertising-market signal may inform an adjustment to a company’s revenue guidance, with the effect supported by explicit calculations. 例如，以外部广告市场信号调整公司的收入指引，并用明确的计算说明这一信号如何影响预测。

946 scored runs; 36.2 primary-model calls per task on average.946次计分运行，平均每题36.2次主模型调用。

Judge评分裁判

02 / the data lake数据湖

Over one million financial documents百万级真实金融文档

The data lake brings together financial information across A-share, Hong Kong and US markets: 10 data categories, including 3.43 million cataloged document versions in nine categories and structured market data. Financial reports, structured statements, earnings calls, investor materials, company announcements, news, research reports, analyst estimates, institutional holdings and stock-market data are updated incrementally.数据湖汇集 A 股、港股与美股金融资料，覆盖10类数据：其中9类目录资料共约343万条文档版本，另有结构化行情数据。财务报告、结构化报表、业绩会记录、投资者关系材料、公司公告与事件、新闻、研究报告、分析师预测、机构持仓及股票行情持续增量更新。

Authorized catalog inventory as of 15 October 2026. Each benchmark task uses the materials permitted by its information cutoff.授权资料目录统计截至 2026 年 10 月 15 日；各评测任务仍按其截止时点限定可用资料。

03 / the freeze冻结

Point-in-time financial information按时点限定的金融信息

The same cutoff applies to searches, reading and downloads. The distribution below covers the 40 tasks with a reference evidence graph.检索、阅读与下载使用相同截止时点；下方统计40道具备参考证据图的题目的文档数量。

Forecasts and subsequent disclosures预测与后续披露

Forecast questions require conclusions based on information available at the task cutoff. Subsequent disclosures provide realised outcomes for comparison. 预测题要求模型依据截止时点前的信息形成判断，后续披露提供可对照的实际结果。

Access controlled by the environment由环境控制数据访问

Each task places the agent at a specific historical moment. Publication times and visibility boundaries determine which documents it can retrieve, excluding later disclosures from the information available for research. 每道题将 Agent 置于一个具体的历史时点。环境依据发布时间与可见时间边界限定可检索文档，研究过程中无法通过工具获取该时点之后的披露。

The same task-level information boundary applies to search, reading and file downloads for all models. 同一题目的检索、阅读和文件下载遵循一致的信息边界，各模型采用相同访问规则。

Recorded evidence and retrieval steps support reproducible historical research. The same point-in-time data environment can also support backtesting and model post-training. 记录下来的证据与检索过程便于复核历史研究；同一套按时点限定的数据环境，也可用于历史回测与模型后训练。

04 / Research environment研究环境

A shared set of research tools统一的研究工具

Each agent uses the same tools to find evidence, analyse financial information and submit an answer. 每个 Agent 使用相同的工具查找证据、分析财务信息并提交答案。

Evidence is limited to each task’s information cutoff. Calculations run in an isolated environment without public internet access. 可用资料受每道题的信息截止时间限制，计算在无公网访问的独立环境中完成。

05 / the tool use工具使用

Tool use across models模型的工具使用情况

Compare searches, reads, downloads, calculations and submissions across each model’s scored runs.比较各模型计分运行中的检索、阅读、下载、计算与提交的工具使用。

Normalisation归一化方式

11 models · field mean on top11 个模型 · 首行为全场均值

06 / Capability differences能力差异

Where model capabilities differ模型能力差异体现在哪里

Research workflow examples illustrate financial formulas, forecast reasoning and tool switching. The comparison below shows mean forecast loss for models and analyst forecasts.研究过程样例展示财务公式、预测逻辑与工具切换的差异；下方比较模型与分析师预测的平均 loss。

From a formula to a financial judgment从公式到金融判断

In the GE Aerospace position-sizing task, Qwen 3.8 Max earns full formula credit but provides no explanation of the protection boundary. A position limit needs a supported equity-downside assumption.在GE Aerospace仓位管理题中，Qwen 3.8 Max拿到了完整的公式分，却没有说明仓位上限对应的风险边界。确定仓位，还需要有依据的股价下行假设。

Disclosed obligations, accounting reserves and equity losses describe different quantities. Connecting them to a position-size decision requires a clear financial argument.披露的担保义务、会计准备和股票损失是不同的量度。把这些数字接到仓位决策上，需要讲清楚相互之间的金融关系。

Reasoning behind a revenue forecast收入预测逻辑的合理性

In the Microsoft pre-earnings sample, Ling-3.0-flash extrapolates from recent results using seasonality and growth assumptions, without anchoring revenue to management’s guidance. GLM-5.3 starts from that midpoint and makes an evidence-based upside adjustment. 微软财报前预测样例中，Ling-3.0-flash 主要依据近期业绩、季节性与增长率假设外推，未以管理层收入指引为锚；GLM-5.3 则在指引基础上，结合证据作出上调。

GLM-5.3 considers prior guidance beats, Azure capacity coming online earlier and Copilot growth, while also accounting for pressure on Windows and gaming. This is closer to an analyst’s evidence-calibrated approach: use guidance as a starting point, then assess whether current business conditions justify a deviation. A higher forecast alone does not make the reasoning stronger. GLM-5.3 同时考虑历史超指引表现、Azure 产能提前交付、Copilot 增长，以及 Windows 和游戏业务的压力。这更接近人类投研专家的分析路径：先以指引为起点，再结合当期经营情况判断是否应偏离指引，而不是仅作机械外推或直接取中值；关键是调整有无依据，而非预测是否更高。

Switching tools to recover evidence取证受阻后的工具切换能力

In the Meta forecast-update sample, the models need Alphabet’s advertising revenue as an external signal. When document reading fails to provide the required figures, GLM-5.3 switches to downloading the original filings and extracting the data locally, reconstructing aggregate advertising growth from Search, YouTube and Network revenue. Meta 预测更新样例需要用 Alphabet 广告收入作为外部信号。阅读工具未返回所需数字时，GLM-5.3 转向下载原始披露并在本地提取数据，从搜索、YouTube 与广告网络收入重建整体广告增速。

Ling-3.0-flash continues searching and reading but does not use the download route in this sample, ultimately relying on an approximate growth figure. GLM-5.3 more effectively turns a blocked reading path into verified source data; Ling-3.0-flash’s response exposes a weakness in adapting its tool strategy. Ling-3.0-flash 则继续搜索与阅读，在这条样例中未使用下载路径，最终采用近似增速。相比之下，GLM-5.3 能更有效地通过工具切换找回并核对原始数据；Ling-3.0-flash 在取证受阻后的策略调整上表现不足。

From forecasts to a trading strategy从预测到交易策略

20 forecast tasks. Compare a model’s forecast with the pre-cutoff analyst reference, then trade the sign of the difference.20 道预测题。比较模型预测与截止日前的分析师预测，再按偏离方向生成交易信号。

Strategy策略

07 / one run, end to end一次完整运行

An agent at work: a recorded research workflowAgent 的真实工作流——以一条轨迹为例

Follow GLM-5.3 as it updates a Tesla forecast, from finding and reviewing sources to calculations and final submission. This independent historical example is not included in the current 86-task aggregates. 以 GLM-5.3 更新特斯拉预测的一次作答为例，展示从资料检索、阅读取证到计算与提交的完整过程；这是独立的历史工作流示例，不混入本轮86题汇总。

Column height柱高口径

Cumulative elapsed time appears below each round每轮下方标注累计耗时

The round-by-round record connects model responses with tool operations. Scoring assesses the final submitted answer; model API costs depend on token usage. 逐轮记录展示模型响应及对应的工具操作。评分依据最终答案，模型 API 费用则按 token 用量计算。

Model and tool time模型调用与工具执行耗时

The 163.6-second research loop includes 127.5 seconds of model calls, 25.0 seconds of tool execution and 11.1 seconds of orchestration overhead. Reading accounts for 16.8 seconds of tool time. Model calls dominate the total duration. 163.6 秒的研究过程中，模型调用占 127.5 秒，工具执行占 25.0 秒，调度等开销占 11.1 秒。其中，阅读工具耗时 16.8 秒。总耗时主要来自模型调用。

Corrections before final submission最终提交前的修正过程

In round 9, the first submit_answer attempt is rejected because the evidence array does not meet the required format. In round 10, the model corrects the fields and resubmits while preserving its research. The two rounds produce 2,422 and 2,017 output tokens respectively. 第 9 轮首次调用 submit_answer 时，evidence 数组不符合规定格式，提交被拒绝。第 10 轮模型保留研究内容，修正字段后重新提交。两轮分别产生 2,422 和 2,017 个输出 token。