Comment by resiros

Are you planning to open-source the benchmark environment and data (even anonymized) to allow people to compete on it. It looks like there are many ways to improve the accuracy of the agent by working on its logic (different tools, multi-agents ...).