DrivenBench puts investment models at 93.9%

DrivenBench’s official leaderboard puts its top two investment-agent models at 93.9%, with each passing 155 of 165 capability checks across real-world workflows; the benchmark page also lists upper-bound costs and median and p90 latency. The third-ranked model costs $160.37 on the benchmark versus $57.69 for a top scorer, so capability rank does not settle procurement. Builders should treat task-specific score, cost, and latency as a bundle; BenchMIRT’s benchmark audit explains why a leaderboard number alone is not deployment evidence.