It's crazy that a lot is happening in open models for stt, but there's very little progress when it comes to results, esp multilingual.