Results

Highlighted: best in its column among the rows shown. n=300 marks a score on part of the test set. ! marks an Uzbek speech score made before our 2026-09-10 scorer fix. Central Asia models are one model for all four languages; models marked three languages left Kyrgyz out.