3.9.0

No App Switching

• Dictate from the keyboard in any app and VivaDicta sends you straight back to where you were typing, instead of leaving you stranded in VivaDicta
• Apple removed the API behind this in iOS 26.4 and every hot mic tap since had ended in the manual return prompt
• The return list also grew by 15 apps, and a limit that was silently breaking the newest entries is gone

Real-Time Dictation in the Keyboard

• The speed boost added in 3.8.0 now works when you dictate from the keyboard, not only inside the app
• Your speech is transcribed while you talk, so the text is ready almost the moment you stop instead of waiting for the recording to upload
• If the connection drops it quietly falls back to the normal upload, so nothing is lost

Speak to Edit

• The keyboard's rewrite tab gains a microphone
• Say what you want done, like "make this more formal" or "shorten to two bullets", and it is applied to your selection, or to the text before the cursor when nothing is selected
• Previously the tab could only apply a preset you had already saved, so anything outside your catalog meant leaving the app you were typing in

More Real-Time Models

• Streaming is no longer one provider: Deepgram Nova 3, Deepgram Flux, ElevenLabs Scribe, and Voxtral Realtime join Soniox Realtime
• Gemini 3.5 Transcribe added, Google's dedicated speech-to-text model with 85+ languages, speaker labels, and hesitation sounds removed as it transcribes
• Each one falls back to a normal upload if its connection drops

Your Notes Can Fill Your Calendar

• Reminder extraction now recognizes calendar events too
• Anything that happens at a set time, like a dinner, a meeting, or an appointment, is offered as an event with its start, end, and location and goes to Apple Calendar
• To-dos keep going to Reminders. Every item is one or the other, never both
• Turn it on with "Extract Calendar Events Too" in Settings

Transcribe With Your Grok Subscription

• A SuperGrok or X Premium sign-in now covers xAI Speech-to-Text as well as Grok chat, so you can transcribe without a console.x.ai key
• If xAI turns the subscription down for speech, it falls back to your API key when you have one rather than failing the recording

Share a Note the Moment It's Ready

• Turn on "Auto Share Note" in Settings and the share sheet opens by itself as soon as a recording finishes transcribing
• Dictating a message to send someone takes one tap instead of four
• Recordings made in the app only, since a keyboard or Watch capture finishes while the app is in the background

Know Which Model to Pick

• The AI Providers screen now opens with the top ten models by measured quality and speed, each one graded on cleaning up real dictation
• Models you can use without paying carry a "Free tier" badge
• The list is collapsed by default so the providers themselves stay on the first screen

Gemini 3.7 Flash

• Google's newest Flash model joins both transcription and AI processing, and becomes the Gemini default
• You can choose how hard Gemini thinks before answering (Low, Medium, or High) and write your own transcription instruction, both in Settings
• Low stays the default because transcription is not a reasoning task, and on a two-minute clip it roughly halves the round trip

Quality of Life

• Notes carrying a tag you mark as excluded now survive auto-delete, keeping their audio as well as their text
• A recording too long for a provider is refused up front with the limit in minutes, instead of a raw error
• Gemini no longer buys a reasoning pass it has no use for on every transcription, roughly halving its round trip
• xAI no longer sends your speech as English when the language is set to auto-detect, so a Russian dictation is transcribed as Russian
• Deepgram Flux writes counts as words again instead of turning "have a good one" into "have a good 1"
• Adding an OpenAI API key no longer fails its check