Very high performance, and much deeper support for tensors than elasticsearch has with with its new dense_vector type. Sadly they both have the same "gotcha" in that it is only really practical to run any _decent_ ranking model as a second phase re-rank to avoid linear scans.