The StarCraft Multi-Agent Challenge (SMAC) is a benchmark for cooperative multi-agent reinforcement learning (MARL). It is based on the real-time strategy game StarCraft II and focuses on micromanagement tasks where each unit is controlled by an independent agent that acts based on local observations. SMAC provides a diverse set of scenarios and recommendations for benchmarking and evaluations. It also includes a deep multi-agent RL learning framework with state-of-the-art algorithms. The goal of SMAC is to provide a standard benchmark environment for years to come, enabling systematic and robust progress in deep MARL. SMAC addresses the challenges of partially observable, cooperative multi-agent learning, where agents must coordinate their behavior while conditioning only on their private observations. This is an attractive research area as such problems are relevant to many real-world systems and are more amenable to evaluation than general-sum problems. SMAC is designed to handle high-dimensional inputs and partial observability, and to learn coordinated behavior even when restricted to fully decentralized execution. SMAC is built on the popular real-time strategy game StarCraft II and uses the SC2LE environment. Instead of tackling the full game of StarCraft with centralized control, SMAC focuses on decentralized micromanagement challenges. Each unit is controlled by an independent, learning agent that acts based only on local observations, while the opponent's units are controlled by the hand-coded built-in StarCraft II AI. SMAC offers a diverse set of scenarios that challenge algorithms to handle high-dimensional inputs and partial observability, and to learn coordinated behavior even when restricted to fully decentralized execution. SMAC is a qualitatively challenging environment that provides together elements of partial observability, challenging dynamics, and high-dimensional observation spaces. It includes a set of StarCraft II micro scenarios that aim to evaluate how well independent agents are able to learn coordination to solve complex tasks. These scenarios are carefully designed to necessitate the learning of one or more micromanagement techniques to defeat the enemy. Each scenario is a confrontation between two armies of units. The goal is to maximise the win rate, i.e., the ratio of games won to games played. The complete list of challenges is presented in Table 1. More specifics on the SMAC scenarios and environment settings can be found in Appendices A.1 and A.2 respectively. SMAC provides a convenient environment for evaluating the effectiveness of MARL algorithms. The simulated StarCraft II environment and carefully designed scenarios require learning rich cooperative behaviours under partial observability, which is a challenging task. The simulated environment also provides an additional state information during training, such as information on all the units on the entire map. This is crucial for facilitating algorithms to take full advantage of the centralized training regime and assessing all aspects of MARL methods. SMAC is a qualitatively challenging environment that provides together elements of partial observability, challenging dynamics, and high-dimensional observation spaces. It includes a set of StarCraft II micro scenarios that aim to evaluate how wellThe StarCraft Multi-Agent Challenge (SMAC) is a benchmark for cooperative multi-agent reinforcement learning (MARL). It is based on the real-time strategy game StarCraft II and focuses on micromanagement tasks where each unit is controlled by an independent agent that acts based on local observations. SMAC provides a diverse set of scenarios and recommendations for benchmarking and evaluations. It also includes a deep multi-agent RL learning framework with state-of-the-art algorithms. The goal of SMAC is to provide a standard benchmark environment for years to come, enabling systematic and robust progress in deep MARL. SMAC addresses the challenges of partially observable, cooperative multi-agent learning, where agents must coordinate their behavior while conditioning only on their private observations. This is an attractive research area as such problems are relevant to many real-world systems and are more amenable to evaluation than general-sum problems. SMAC is designed to handle high-dimensional inputs and partial observability, and to learn coordinated behavior even when restricted to fully decentralized execution. SMAC is built on the popular real-time strategy game StarCraft II and uses the SC2LE environment. Instead of tackling the full game of StarCraft with centralized control, SMAC focuses on decentralized micromanagement challenges. Each unit is controlled by an independent, learning agent that acts based only on local observations, while the opponent's units are controlled by the hand-coded built-in StarCraft II AI. SMAC offers a diverse set of scenarios that challenge algorithms to handle high-dimensional inputs and partial observability, and to learn coordinated behavior even when restricted to fully decentralized execution. SMAC is a qualitatively challenging environment that provides together elements of partial observability, challenging dynamics, and high-dimensional observation spaces. It includes a set of StarCraft II micro scenarios that aim to evaluate how well independent agents are able to learn coordination to solve complex tasks. These scenarios are carefully designed to necessitate the learning of one or more micromanagement techniques to defeat the enemy. Each scenario is a confrontation between two armies of units. The goal is to maximise the win rate, i.e., the ratio of games won to games played. The complete list of challenges is presented in Table 1. More specifics on the SMAC scenarios and environment settings can be found in Appendices A.1 and A.2 respectively. SMAC provides a convenient environment for evaluating the effectiveness of MARL algorithms. The simulated StarCraft II environment and carefully designed scenarios require learning rich cooperative behaviours under partial observability, which is a challenging task. The simulated environment also provides an additional state information during training, such as information on all the units on the entire map. This is crucial for facilitating algorithms to take full advantage of the centralized training regime and assessing all aspects of MARL methods. SMAC is a qualitatively challenging environment that provides together elements of partial observability, challenging dynamics, and high-dimensional observation spaces. It includes a set of StarCraft II micro scenarios that aim to evaluate how well