QuantBench financial reasoning benchmark
QuantBench evaluates whether language models can reason about financial decisions, obey production contracts, and avoid unsafe approvals across 106 cases.
Latest public run
106 cases, 16 scoring categories, and 15 evaluated model routes. The leaderboard reports overall score, JSON compliance, action accuracy, recorded provider cost, token use, false approvals, missed approvals, and HTTP or parsing failures.
Download the results dataset · Inspect provenance
Interpretation and limitations
QuantBench measures financial reasoning, production-contract compliance, and capital-safety decisions under a fixed case bank. It is evidence about benchmark performance, not a trading guarantee. Most hosted entries are single runs, provider routes and model revisions can change, and local versus hosted infrastructure is not directly comparable.