MolGAN is an implicit, likelihood-free generative model designed for small molecular graphs. It leverages generative adversarial networks (GANs) to directly generate molecular graphs, avoiding the need for expensive graph matching procedures or node ordering heuristics. The model combines GANs with a reinforcement learning (RL) objective to encourage the generation of molecules with specific desired chemical properties. Experiments on the QM9 chemical database demonstrate that MolGAN can generate nearly 100% valid compounds, outperforming both string-based representations and likelihood-based methods. The model's effectiveness is attributed to its ability to implicitly optimize valid molecules and its use of a deterministic policy gradient algorithm, which helps in generating diverse and chemically valid compounds. However, the model is susceptible to mode collapse, where it tends to generate a limited set of diverse outputs, which is a limitation that future work aims to address.MolGAN is an implicit, likelihood-free generative model designed for small molecular graphs. It leverages generative adversarial networks (GANs) to directly generate molecular graphs, avoiding the need for expensive graph matching procedures or node ordering heuristics. The model combines GANs with a reinforcement learning (RL) objective to encourage the generation of molecules with specific desired chemical properties. Experiments on the QM9 chemical database demonstrate that MolGAN can generate nearly 100% valid compounds, outperforming both string-based representations and likelihood-based methods. The model's effectiveness is attributed to its ability to implicitly optimize valid molecules and its use of a deterministic policy gradient algorithm, which helps in generating diverse and chemically valid compounds. However, the model is susceptible to mode collapse, where it tends to generate a limited set of diverse outputs, which is a limitation that future work aims to address.