Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?
The top-scoring agent managed just 52.5 on forecasting and modeling despite leading the benchmark overall.
Investor’s Last Exam tests equity research using information frozen at the time of each decision, with timestamp controls intended to prevent access to future evidence. Built under experienced buy-side investors and fund managers, it covers 86 assignments and assesses research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra led overall with 68.1 out of 100, followed by Claude Opus 5.5 at 66.9. The evaluation traces sources, calculations, and assumptions to identify where research capability fails to translate into forecasts and investment judgments.
HF Daily Papers' note
Investor’s Last Exam tests equity research using information frozen at the time of each decision, with timestamp controls intended to prevent access to future evidence. Built under experienced buy-side investors and fund managers, it covers 86 assignments and assesses research quality, forecast accuracy, and simulated investment performance. GPT-6 Astra led overall with 68.1 out of 100, followed by Claude Opus 5.5 at 66.9. The evaluation traces sources, calculations, and assumptions to identify where research capability fails to translate into forecasts and investment judgments.
HF Daily Papers' note
score 5