Macaw is an AI model designed to answer a broad range of questions. It has been trained in “multi-angle” fashion, meaning it can process multiple types of inputs and outputs (such as questions, answers, and explanations). Macaw comes in different sizes—macaw-11b, macaw-3b, macaw-large—as well as an answer-focused version that’s featured on various leaderboards: macaw-answer-11b. Compared to GPT-3 (which uses 175 billion parameters), Macaw only requires 11 billion parameters yet outperforms the popular model. This makes Macaw a powerful tool for answering general knowledge questions, meta reasoning queries, hypothetical scenarios and story understanding tasks. Developed by AI2’s Aristo team as part of their mission to create systems that read, reason and explain their answers accurately and efficiently