Market · AI in practice
DeepFinance helps measure AI on complex financial analysis
Julius Research published a test with 21 tasks based on analyses used in investments and operations; the top published score was 54.8% against the evaluation criteria, with tasks averaging 27.1 human hours each.
What the sources report
On November 5, 2026, Julius Research published DeepFinance, a test of how AI models handle financial analyses used in investments and operations.
The set has 21 tasks. Each one averages 27.1 hours of human work. In the published result, the top model scored 54.8% against the evaluation criteria.
Adriano's reading
Those numbers help compare systems on work that often takes days of human attention. They also give a shared reference for teams building finance products.
What interests me is having a concrete way to measure AI on this kind of analysis. It gives teams a starting point for building and improving agents. I see an opportunity to make this kind of analysis more accessible and give professionals more time to explore investment decisions.