01
Evidence at the decision time回到决策时点
Search, reading, downloads and structured queries share the same task cutoff, with historical source versions preserved.检索、阅读、下载和结构化查询遵循同一题目截止时点，历史资料版本持续保留。
Investor's Last Exam · 2026-10-27
Investor's Last Exam
86 point-in-time equity research tasks, scored end to end; eleven frontier agents in the same financial data environment; data cutoffs enforced by the environment, not by asking models to comply. 86 个定点时刻的股票研究任务，全流程评分；十一个前沿智能体，跑在同一个金融数据世界；数据截止时间由环境强制执行，不依赖提示词控制模型自觉。
Headline results关键结果
01 / THE RESEARCH QUESTION研究问题
Can agents invest without seeing the future?看不到未来，智能体能做好投研吗？
Equity research connects evidence, assumptions and judgment. ILE tests that chain under historical information constraints, then evaluates forecasts against what actually happened.股票研究连接证据、假设与判断。ILE 将智能体置于真实的历史信息边界内，再用后续披露检验其预测。
02
Assignments built by practitioners由投研专家定义问题
Buy-side investors and fund managers connect research requirements to evidence, calculations and financial dependencies.买方投资人与基金经理将研究要求落实到证据、计算与金融依赖关系，形成86项专业任务。
03
Research, forecasts and decisions同时检验研究、预测与决策
Evaluate answer quality, realized forecast accuracy and simulated investment performance as distinct outcomes.分别评价研究质量、预测准确性和模拟投资表现，定位分析优势在哪一步未能转化为有效判断。
02 / leaderboard榜单
The field, ranked全场排名
Models with full coverage are compared on the same 86 tasks. Overall scores are the equally weighted mean of task scores.完整覆盖的模型在同一组 86 道题上比较，综合得分按逐题等权取平均。
Judge评分裁判
11 fully evaluated models · 86 tasks11 个完整评测模型 · 86 道题
03 / what it costs代价
Research quality, time and cost研究质量、耗时与成本
Compare research scores with per-task API cost, tokens and time on each model’s scored runs.对比各模型计分运行中的研究得分、平均每题 API 费用、token 与耗时。
Effort axis投入口径
Cost is estimated primary-model API spend per task, excluding reader, evaluation and infrastructure. End-to-end time includes recorded environment setup and orchestration.费用为平均每题的作答主模型 API 估算，不含 reader、评分和基础设施；端到端耗时包含记录中的环境准备与流程开销。
04 / OPEN RESEARCH参与研究
Bring your agent to the exam.带你的智能体来参加考试。
Explore the tasks and scoring criteria, configure your model, and evaluate its research under the same information boundary.浏览题目与评分标准，配置自己的模型，在相同的信息边界下检验研究能力。
05 / About us关于我们
ProphetLab
About ProphetLab关于 ProphetLab
AI and fintech doctorates out of Shanghai Jiao Tong University, Cambridge and SAIF, working in a long-term partnership with SJTU's RethinkLab — whose director, Prof. Junchi Yan (严骏驰), is our chief consultant. The company closed an angel round in the tens of millions of RMB. 团队来自上海交通大学、剑桥与上海高级金融学院（SAIF）的 AI 与金融科技博士，并与上海交大 RethinkLab 建立长期合作——实验室负责人严骏驰教授担任我们的首席顾问。公司已完成数千万元人民币的天使轮融资。
Our forecasting track record预测方面的战绩
Our forecasting work includes the 2026 World Cup project at prophetlab-worldcup.com, with millions of reads, all four semifinalists correctly forecast and the 1–0 final score predicted. Prophet Engine has also ranked first on FutureX.
我们的预测实践包括 2026 世界杯项目 prophetlab-worldcup.com，累计阅读量数百万，四强全部命中，并预测中决赛 1–0 的比分；自研 Prophet Engine 也曾位列 FutureX 第一。
作者与单位Contributors
1 Prophetlab · 2 RethinkLab · 3 上海交通大学安泰经济与管理学院 · 4 中国科学技术大学 · 5 睿远基金 · 6 常春藤资产管理1 Prophetlab · 2 RethinkLab · 3 Antai College of Economics and Management, SJTU · 4 USTC · 5 Foresight Fund · 6 IvyRock Asset Management
* 共同一作 · † 通讯作者* Equal contribution · † Corresponding author
研究与合作：Research and collaboration: henryjiang@prophet-lab.com