Wispr Flow

Recently shipped a 30% reduction in dictation latency since January and a fix for an overly aggressive Auto Cleanup setting that was causing accuracy problems.

The Flow Bar is now draggable. You can reposition it to the left or right edge instead of being stuck at the bottom of the screen, and it remembers your position per session. Canadian English speakers also get a one-time prompt to switch locales for correct spellings.

iOS dictation interruptions no longer lose your work. A notification confirms the transcript is saved when a call, Siri, or the five-minute limit cuts in. Tapping opens the transcript with auto-copy, and failed dictations get a one-tap retry button. The Live Dictation Timer now runs in the Dynamic Island for every dictation, and the keyboard backspace works like the iOS system one with word-level acceleration.

Android reliability was rebuilt from the ground up. The Flow Bubble and accessibility permission now self-recover if lost, multi-account settings stay isolated, and Xiaomi/OnePlus battery optimization conflicts were resolved. Crash fixes cleaned up the remaining rough edges.

More apps gained auto-switchback support on iOS. Works across every supported iOS version including the iOS 27 beta, so you're never stuck manually tapping back to Claude, ChatGPT, or LinkedIn after dictating.

Accuracy and formatting regained consistency. All dictations now route through the same current model, eliminating session-to-session variation and fixing course corrections that didn't land, swapped words, and ignored punctuation. A/B testing and expanded benchmarks now catch model regressions before shipping.

Desktop localization expanded and new data controls arrived. The Flow Hub is now available in English, German, Spanish, Italian, and Portuguese. A new Cloud Sync setting alongside Privacy Mode controls whether transcription data is stored on servers for cross-device sync, separate from model training use. Infrastructure was scaled up and demanding work isolated to dedicated systems, improving dictation speed, accuracy, and sign-in reliability.