Active learning (AL) algorithms may achieve better performance with fewer\ndata because the model guides the data selection process. While many algorithms\nhave been proposed, there is little study on what the optimal AL algorithm\nlooks like, which would help researchers understand where their models fall\nshort and iterate on the design. In this paper, we present a simulated\nannealing algorithm to search for this optimal oracle and analyze it for\nseveral tasks. We present qualitative and quantitative insights into the\nbehaviors of this oracle, comparing and contrasting them with those of various\nheuristics. Moreover, we are able to consistently improve the heuristics using\none particular insight. We hope that our findings can better inform future\nactive learning research. The code is available at\nhttps://github.com/YilunZhou/optimal-active-learning.\n