Bi-DexHands provides a collection of bimanual dexterous manipulation simulation tasks integrated with reinforcement learning algorithms. It leverages NVIDIA Isaac Gym to enable running thousands of parallel environments on a single GPU, significantly accelerating RL training. The library supports both visual and low-level state observations, integrates with rl-games for training, and is used for research into achieving human-level hand dexterity and bimanual coordination in robotic systems.