GTAGym: Real AAA Game Tasks Environment for Evaluation and Reinforcement Learning

GTAGym is a new environment and framework for evaluating and applying reinforcement learning to GTA V tasks. Its GTAVBench benchmark includes 103 online tasks from GTA V main-story missions across nine task families and 47 image-based shooting and sniping tasks. A web server supports parallel testing via screenshots and keyboard-mouse actions with automatic engine-state outcome checks. Six frontier LLMs were evaluated on GTAVBench and its GTAVBench-AGI branch, with the best configuration achieving 70.9% overall accuracy in GTAVBench-AGI.