Improvement
Feedback lands faster after you speak
Corrections now arrive while the sentence is still fresh in your ear, the moment they’re most useful.
Corrections used to take too long after you finished speaking, long enough that the sentence had already faded from your ear. We rebuilt how the model returns feedback so it comes back while the exercise is still fresh.
Nothing about the feedback itself changed; it’s the same specific, native-speaker-level correction on your actual output. It just arrives when it’s most useful: the moment you’re still holding the sentence in your head.
Related What’s New