Investor's Last Exam tests equity agents on frozen decision-time evidence
Original title: Investor's Last Exam: Can AI Agents Do Equity Investment Without Seeing the Future?
GPT-6 Astra leads at 68.1 of 100, but scores only 52.5 on Forecast & Model.
Researchers introduce Investor's Last Exam, an equity-research benchmark built with buy-side investors and fund managers. It uses a versioned database of 3.43 million document versions across nine source categories, timestamped retrieval, and expert scoring tied to financial dependency graphs. Across 86 assignments, GPT-6 Astra and Claude Opus 5.5 lead at 68.1 and 66.9, yet the authors say strong research scores do not become strong forecasts.
- The database spans 3.43 million document versions and nine source categories.
- Scoring covers 86 assignments on research quality, forecasts, and simulated performance.
- GPT-6 Astra scores 68.1 overall and 52.5 on Forecast & Model.
- Claude Opus 5.5 scores 66.9 overall.
- AI models
- GPT-6 AstraClaude Opus 5.5
From the original article
1 min read at huggingface.co
Financial agents must form forecasts and investment judgments from decision-time evidence, with predictions evaluated against subsequent outcomes. Existing benchmarks assess financial knowledge, research workflows, or forecasting, but evaluating the full research process requires a strictly frozen information set and professional evaluation criteria.
By Huangfei Jiang, Anning Hu, Kai Sheng · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.