# Authors introduce a benchmark for testing AI agents’ equity research and investment judgments [Canonical reading](https://escapemeridian.com/readings/authors-introduce-a-benchmark-for-testing-ai-agents-equity-75b5c455) The authors introduce Investor's Last Exam (ILE), a benchmark designed to test whether financial agents can make investment judgments using only information available at decision time. They argue that evaluating financial agents requires a frozen information set and professional criteria for the full research process, rather than tests of financial knowledge, research workflows, or forecasting alone. ILE combines a versioned financial database with structured market data, tasks organized around analyst deliverables, and expert scoring that links judgments to sources, calculations, and assumptions. Across 86 assignments, the authors assess research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra and Claude Opus 5.5 lead overall, but the authors report that even Astra scores only 52.5 out of 100 on Forecast & Model. The authors say their results show that even state-of-the-art models fail to translate overall research capability into predictive performance. ILE is designed to trace the chain from evidence through assumptions to judgment, helping researchers locate where analytical advantages fail to translate into forecasts and investment judgments. The abstract reports simulated investment performance but does not establish how benchmark results correspond to real-world investment outcomes. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source arXiv cs.AI · Article Original source date: Oct 28, 2026 (2026-10-28). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T13:24:04.992Z. First observed: 2026-10-30T06:42:53.429Z. Retrieved: 2026-10-31T07:33:12.427Z. - [Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?](https://arxiv.org/abs/2610.38342) ## Limits Evidence is the paper’s arXiv abstract; it does not include the full methods, assignments, scoring details, or results beyond those summarized there. The abstract reports simulated investment performance but does not describe its relationship to real-world investment outcomes. ## Open question How does ILE’s expert-led framework locate where analytical advantages fail to translate into forecasts and investment judgments? ## Source-stated The authors introduce ILE as an equity-research benchmark using decision-time evidence and professional evaluation criteria. > We introduce Investor's Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated ILE combines a versioned financial database, structured market data, analyst-deliverable tasks, and an expert-led scoring framework linking judgments to sources, calculations, and assumptions. > Its three components are a versioned financial database spanning 3.43 million document versions and nine source categories, supplemented by structured market data - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. > Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The authors report that GPT-6 Astra and Claude Opus 5.5 lead overall, while Astra scores 52.5 on Forecast & Model. > GPT-6 Astra and Claude Opus 5.5 lead overall at 68.1 and 66.9 out of 100, yet even Astra scores only 52.5 on Forecast & Model. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The authors argue that leading overall research capability does not necessarily translate into predictive performance. > Our results show that even state-of-the-art models fail to translate overall research capability into predictive performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Escape Meridian interpretation The abstract does not establish how benchmark scores correspond to real-world investment outcomes. > Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source and contributor context Jiang, Huangfei, Hu, Anning, Sheng, Kai, Cai, Yuchen, Ge, Rui, Kong, Xiangyu, Lin, Jianghao, Chen, Shuzi, Weng, Jiyi, Yan, Junchi · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: d467821f-95e4-46c6-8a85-1660845fd687. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T13:24:04.992Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 73851ce35c562cfb72d36e70bb5de4ece30bc0c6b4cf3207b8928d14b7f4522d. Source text hash: e5034e6af6942c45d0e228b759e6f8ac611f7ba3dbf8638b510653670855447b. ## Revision and source-check history ### Version 1 — 2026-11-11T13:24:04.992Z The authors introduce Investor's Last Exam (ILE), a benchmark designed to test whether financial agents can make investment judgments using only information available at decision time. They argue that evaluating financial agents requires a frozen information set and professional criteria for the full research process, rather than tests of financial knowledge, research workflows, or forecasting alone. ILE combines a versioned financial database with structured market data, tasks organized around analyst deliverables, and expert scoring that links judgments to sources, calculations, and assumptions. Across 86 assignments, the authors assess research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra and Claude Opus 5.5 lead overall, but the authors report that even Astra scores only 52.5 out of 100 on Forecast & Model. The authors say their results show that even state-of-the-art models fail to translate overall research capability into predictive performance. ILE is designed to trace the chain from evidence through assumptions to judgment, helping researchers locate where analytical advantages fail to translate into forecasts and investment judgments. The abstract reports simulated investment performance but does not establish how benchmark results correspond to real-world investment outcomes. ## Original source arXiv cs.AI · Article Original source date: Oct 28, 2026 (2026-10-28). Date basis: Date basis not provided. Added to Escape Meridian: 2026-11-11T13:24:04.992Z. First observed: 2026-10-30T06:42:53.429Z. Retrieved: 2026-10-31T07:33:12.427Z. - [Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?](https://arxiv.org/abs/2610.38342) ## Limits Evidence is the paper’s arXiv abstract; it does not include the full methods, assignments, scoring details, or results beyond those summarized there. The abstract reports simulated investment performance but does not describe its relationship to real-world investment outcomes. ## Open question How does ILE’s expert-led framework locate where analytical advantages fail to translate into forecasts and investment judgments? ## Source-stated The authors introduce ILE as an equity-research benchmark using decision-time evidence and professional evaluation criteria. > We introduce Investor's Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated ILE combines a versioned financial database, structured market data, analyst-deliverable tasks, and an expert-led scoring framework linking judgments to sources, calculations, and assumptions. > Its three components are a versioned financial database spanning 3.43 million document versions and nine source categories, supplemented by structured market data - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. > Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The authors report that GPT-6 Astra and Claude Opus 5.5 lead overall, while Astra scores 52.5 on Forecast & Model. > GPT-6 Astra and Claude Opus 5.5 lead overall at 68.1 and 66.9 out of 100, yet even Astra scores only 52.5 on Forecast & Model. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source-stated The authors argue that leading overall research capability does not necessarily translate into predictive performance. > Our results show that even state-of-the-art models fail to translate overall research capability into predictive performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Escape Meridian interpretation The abstract does not establish how benchmark scores correspond to real-world investment outcomes. > Evaluation spans 86 assignments and jointly assesses research quality, forecast accuracy, and simulated investment performance. - [Supporting source](https://arxiv.org/abs/2610.38342) ## Source and contributor context Jiang, Huangfei, Hu, Anning, Sheng, Kai, Cai, Yuchen, Ge, Rui, Kong, Xiangyu, Lin, Jianghao, Chen, Shuzi, Weng, Jiyi, Yan, Junchi · Article author These are credits stated by the source; they do not establish a profile match or authorship of every passage. ## Source media No suitable media was found by the supported page extraction. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## AI agent An AI system that chooses and carries out actions toward a goal, often using software tools. - [Definition reference](https://www.ibm.com/think/topics/ai-agents) ## arXiv arXiv is a research manuscript repository where authors share papers that may not yet have been peer reviewed. - [Explore this context](/entities/arxiv) - [Reference](https://arxiv.org/) - [Reference](https://info.arxiv.org/help/api/index.html) ## TIME TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services. - [Explore this context](/entities/time) - [Reference](https://time.com/) ## Claude Claude is Anthropic’s family of AI models and assistants for tasks such as writing, analysis and coding. - [Explore this context](/entities/claude) - [Reference](https://claude.com/product/overview) - [Reference](https://www.anthropic.com/news) - [Reference](https://www.anthropic.com/research/team/economics) ## Editorial provenance Version 1. Publication ID: d467821f-95e4-46c6-8a85-1660845fd687. Prepared by gpt-6-luna; grounding review by gpt-6-luna. Source check: passed by anthropic/claude-opus-5.5 on 2026-11-11T13:24:04.992Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: material\_development. Observation: 73851ce35c562cfb72d36e70bb5de4ece30bc0c6b4cf3207b8928d14b7f4522d. Source text hash: e5034e6af6942c45d0e228b759e6f8ac611f7ba3dbf8638b510653670855447b. ### passed — 2026-11-11T13:24:04.992Z — anthropic/claude-opus-5.5 Publication: d467821f-95e4-46c6-8a85-1660845fd687 attribution — pass: The reading consistently attributes ILE, the evaluation design, the model scores and the conclusion about predictive performance to the paper's authors. It does not present their findings as established fact or as Escape Meridian's view. > We introduce Investor's Last Exam (ILE), an equity-research benchmark led by experienced buy-side investors and fund managers. summary — pass: The title, summary and question faithfully convey the core argument: AI agents should be judged on investment research using only decision-time information, and top models' overall research ability does not carry over to forecasting. The phrase 'only information available at decision time' makes the not-seeing-the-future idea clear to general readers. Two optional fidelity repairs: state the overall scores (68.1 and 66.9) to sharpen the contrast with 52.5, and note that validated timestamps govern retrieval. The question is grounded in the source. > Our results show that even state-of-the-art models fail to translate overall research capability into predictive performance. date — pass: The originalPublished date of 2026-10-28 matches the arXiv v1 submission date. The later feed date of 2026-10-31 is correctly not used. > \[Submitted on 28 Oct 2026\] support — pass: Each claim is supported by a verbatim abstract excerpt. The interpretation that the abstract does not establish a link to real-world outcomes is labeled as interpretation and is consistent with the source's reference to simulated performance. The reading adds no unsupported benefits, costs or forecasts. 'Links judgments' slightly paraphrases 'linking scoring criteria', but this does not materially change the meaning. > an expert-led framework linking scoring criteria to sources, calculations, and assumptions through financial dependency graphs media — pass: The selected media is the verified direct link to the arXiv abstract page. That page is the original source. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/75b5c455290653e020549e2a711608202331fd42d197f202bb11a0e06948e369/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/authors-introduce-a-benchmark-for-testing-ai-agents-equity-75b5c455.json)