Skip to content Black FridayUp to 76% off Vitals, Levers and SoundBar, until 5 December
Flit · Languages and models

Dictate in 1,600+ languages with on-device voice models

Flit runs every voice model on your Mac and picks the right one for the language it hears. Thirty languages appear word by word while you speak. The rest appear when you stop, and all of it works offline.

Language chips from English and German to Hindi and Ukrainian, ending in “+1,600 more”, and a sentence said in German, “können wir uns äh am freitag um halb elf treffen”, written in English as “Can we meet on Friday at 10:30?”

It hears which language you are speaking

Tell Flit which languages you dictate in, in the welcome window or in Settings → Models. From then on it identifies the language of each dictation in the first seconds and switches engine when the current one cannot handle it, so a bilingual day does not mean trips to Settings. Detection uses a 75 MB model that is downloaded once and runs on the Mac. When a dictation is too short to tell, Flit falls back to the first language on your list.

Which voice model for which language

You only need one voice model, and Flit recommends one from your languages and your Mac’s memory. Whisper is there in several sizes too, and Apple’s own speech recognition needs no download at all. Every model shows its size and can be deleted and fetched again.

Parakeet Ultra is Parakeet V3 trained further. Across 24 languages of the FLEURS test set its average word error rate is 11.7%, against 14.8% for V3, at the same speed. Voxtral shows words a second or two behind your voice. Every model works for recordings in the Transcribe tab as well.

  • Parakeet Ultra 0.6B: English and 24 more European languages, live words, 595 MB
  • Cohere Transcribe: the most accurate for German, French, Spanish, Italian, Dutch and nine more; text appears when you stop
  • Voxtral Realtime 4B: live words in 13 languages, Hindi, Arabic, Japanese, Korean and Chinese among them; 2.8 GB
  • Qwen3-ASR: 30 languages, such as Hindi, Turkish, Thai and Indonesian
  • Meta Omnilingual ASR 3B: more than 1,600 languages; 1.8 GB
  • Parakeet Unified 0.6B: English only, live words, 570 MB
The model Library in Flit’s settings: Parakeet Ultra 0.6B marked Active, then Cohere Transcribe, Qwen3-ASR, Voxtral Realtime 4B, Meta Omnilingual ASR 3B and Parakeet Unified 0.6B, each with its speed, accuracy and size

The widest coverage, and its limits

Omnilingual is what takes the count past 1,600. It covers languages no other model in Flit does, among them Tamil, Telugu, Bengali, Marathi, Urdu, Punjabi, Swahili, Yoruba and Amharic. It writes plain words with no punctuation or capitals, and Flit’s formatting adds them at the Polished level.

Accuracy depends on the language. In Flit’s tests it was close to word for word in Hindi, made a few slips in Arabic, and more in Tamil and Bengali. Where another model covers your language, that model is the better choice.

The voice model step of Flit’s setup for someone who speaks Tamil: Omnilingual marked Recommended, beside Parakeet Ultra and Cohere Transcribe

Speak one language, write another

Translation is separate and off by default. Turn it on and you can dictate in German and get English: “können wir uns äh am Freitag um halb elf treffen” lands as “Can we meet on Friday at 10:30?”. Nothing is translated unless you ask.

A Mail reply to “Passt dir Freitag um halb elf?” that reads “Can we meet on Friday at 10:30?”, tagged spoken in German, written in English

The app’s own language

Since 1.7 the app itself, from setup to Settings, comes in 11 languages: English, German, French, Spanish, Italian, Swedish, Dutch, Brazilian Portuguese, Japanese, Chinese and Korean. It follows the language of your Mac, and that does not limit what you can dictate. Spoken commands are said in English for now, whatever language you dictate in.

How to use it

  1. 1

    Open Settings → Models

    Right-click the Flit icon in the menu bar and choose Settings…, then Models in the sidebar.

  2. 2

    Download a model

    Find it in the Library and click Download. The model you use now keeps working meanwhile.

  3. 3

    Make it active

    When the row says Downloaded, click it. It is marked Active.

  4. 4

    Dictate as usual

    The first dictation after a switch takes a few seconds longer while the model loads.

The details

Languages
More than 1,600, all on your Mac
Live as you speak
30: Parakeet’s 25 European languages, plus Chinese, Japanese, Arabic, Hindi and Korean
Language detection
Automatic, with a 75 MB model
Model sizes
Parakeet Unified 570 MB · Parakeet Ultra 595 MB · Omnilingual 1.8 GB · Voxtral Realtime 2.8 GB
Memory
Voxtral Realtime and Cohere Transcribe suit 16 GB or more
Translation
Off by default
App interface
11 languages

Questions

Do I have to switch language before I speak?

No. Add your languages once and Flit detects the language of each dictation.

Which voice model is best for my language?

Open Settings → Models and add your languages under Languages you speak. The card at the top names the model Flit recommends, and why.

Do I need to download every model?

No. One is enough, and you can delete the ones you do not use.

Does Voxtral run on a Mac with 8 GB of memory?

It is built for 16 GB or more. On 8 GB, Parakeet Ultra or Qwen3-ASR 0.6B is the better choice.

Do the voice models need the internet?

Only for the download you start yourself. After that every model runs on your Mac, with Wi-Fi off if you like.

The same feature on the app's own site: flit.fyi/changelog/1.6

Stop typing what you could just say.