Transkribus is unusual in this lineup because it is designed to be trained on your specific archive rather than used off the shelf. Untrained, it returned 47.7% WER on our sample, with most words wrong, which is exactly what its own documentation would lead you to expect.
The pay-off comes after investment. For a large collection in a single consistent hand, say hundreds of pages of one person's letters, training a model can reach strong accuracy, and the platform has deep roots in the academic and archival world.
For everyday handwriting, or any project where you cannot spend 50 or more hours labelling training data, it is not competitive with a specialist that works on the first upload. Match the tool to the project: Transkribus for funded, single-hand archival work; a ready-to-use specialist for everything else.