OpenFinGym: Unified Multi-Task Environment for Quant Agent Evaluation.

A new unified gym environment called OpenFinGym has been developed to address the fragmented evaluation landscape for quantitative-finance agents. While large language model agents are increasingly applied to financial workflows, existing platforms often focus on isolated tasks, potentially overstating agent competence and failing to reveal weaknesses in generalization or real-market interaction. OpenFinGym aims to provide a comprehensive solution by covering interdependent tasks such as forecasting, strategy construction, risk management, and trading within a single interface. OpenFinGym offers several key features to facilitate robust agent development and evaluation. It includes an automated task-construction pipeline that converts quantitative finance publications into executable task packages, streamlining the creation of relevant benchmarks. The platform also provides a containerized runtime with a host-side verifier service, ensuring scalable agent rollouts and preventing train-test leakage. Further enhancing its utility, OpenFinGym incorporates a paper trading engine with a low-latency data-stream design, support for deferred-resolution in long-horizon forecasts, and integration for supervised fine-tuning (SFT) and reinforcement learning (RL) post-training. This holistic environment allows for a more realistic and verifiable assessment of quant agents across multi-stage financial workflows.