One of the challenging problems in sequence generation tasks is the optimized\ngeneration of sequences with specific desired goals. Current sequential\ngenerative models mainly generate sequences to closely mimic the training data,\nwithout direct optimization of desired goals or properties specific to the\ntask. We introduce OptiGAN, a generative model that incorporates both\nGenerative Adversarial Networks (GAN) and Reinforcement Learning (RL) to\noptimize desired goal scores using policy gradients. We apply our model to text\nand real-valued sequence generation, where our model is able to achieve higher\ndesired scores out-performing GAN and RL baselines, while not sacrificing\noutput sample diversity.\n