FinanceBench
Financial document QA; 150 open cases of 10,231; launch paper GPT-4-Turbo with retrieval failed or refused about 81%
domain
Definition
- Version
- not in the notes
- Category
- domain
- Maintainer
- Patronus AI
- Status
- not in the notes
- Home page
- not in the notes
Sources
- FinanceBench: 81% of Answers Were Wrong capitalandcompute.net https://capitalandcompute.net/ai-benchmarks/financebench/
Scores
| Model | Score | Who ran it | Setting | View or population | Date | Source |
|---|---|---|---|---|---|---|
 GPT-5.5 | 91 % | vendor-reported | Aggregator score, not comparable to paper setup; not independently confirmed | rank 1, – | 14 Aug 2026 | capitalandcompute.net |
Where this comes from
Scores from different settings, harnesses or dates are not directly comparable; the row says which. Sorted by score, highest first, as a reading aid, not a ranking.
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "FinanceBench", https://aurion.semai.id/benchmarks/financebench/, records of October 2026.