Introducing Gemini 3.5 Transcribe 🚀 Our most precise speech to text model, that can handle over 85+ languages, has smart correction, and custom vocabulary. I vibe coded a Wispr Flow like app powered by the model. Demo + open sourcing below!

Make your own dictation apps using this repo!

how good is it for realtime tho? like showing things as ur saying it?

There’s a -live streaming variant of the model! I’ll actually add that to the repo and reply with a clip once I get a moment :)

This is expected to be good, trying it today

lmk what you think!

Ammaar, two questions if I may: 1) Can we set the model to do verbatim transcriptions only? I'm transcribing court recordings, and correcting multiple sentences into one ruins the precision I need. 2) Will you release the individual locale data for this chart (e.g., in a whitepaper)?

1. Yes you can switch to a verbatim mode! 2. I can’t speak to as well as others on the team!

damn and he open sourced it 😮‍💨 💥

Didn't @WisprFlow say that Google was coming for them? And didn't @OfficialLoganK say no one is coming for them, in this response that is now deleted? @tankots what do you think? 👀

Jot is my favorite new app for voice notes. Up next - calendar and Docs integration :)

Wispr Flow: Google is coming after us Google: Don’t worry Wispr, we are not Google Dev: check out my vibe coded Wispr replacement app

finally google employees got something to tweet after series of vague posting

🔥🔥🔥

the smart correction alone is a massive win for niche terminology, nice work

its my favorite feature! makes rambling way more accurate.

helps my rambling too

does custom vocabulary mean it finally stops mangling names on the first pass or is that still aspirational

@GabiiAH11 Yes you can specify names and it’ll remember them!

looks amazing, congrats. is it possible to see the transcribed text as I speak?

alright, it's alive!

Thanks! This is awesome

I just love that Wisper Flow is getting slammed daily

Not your app’s fault but would be nice if Google didn’t rate limit me like idiots. I dictated one email and boom - rate limited. Didn’t even finish it.

Can you send me more details, can share with the team and help get you setup!

is the model open source ? how much does it cost to run ??

That desktop background vibe is so clean too!

zen state!

i'm relocating to that state!

What's the pricing for the api . Couldn't find any info regarding it.

the fact you can build your own Wispr Flow-style app on top of this is probably the coolest part. speech is becoming an input layer developers can add to almost anything.

serial killer-level organization with only 2 notes listed in Notes app

Ahahahahaha new laptop without notes sync turned on yet. I assure you, my notes are absolutely chaotic 😂

@ammaar amazing job! I made some small adaption and made a version that only use Voxtral model with realtime transcription all LOCAL and FREE. Amazingly fast and also 100% secure. Thanks for the nice UI work on the app side. Github repo:

Can it detect other languages would love the demo that as well

@grok how cheap is this vs OpenAI

The latency is too high

👏👏👏👏

Now please make it power the Gboard on Android 🙏

Bravo it was needed intime

Congrats. I just opened this. I like that you vibe coded the Wispr clone in the same post.

Why so expensive, how is it better than 3.5 flash lite?

Alot was really on deck for this to happen

Will it be used for Gemini live? Currently speech transcription is pretty bad even though the model understands the audio well!

Just added it to @soniox_ai Compare tool, so everybody can try it side-by-side with other major STT models

It not normal that you limite file to only hour of from file upload !