Fuck You FlowHome / Guides / Which Whisper Model Should I Use? A Plain Guide
Which Whisper Model Should I Use? A Plain Guide
From version 0.9.10 Fuck You Flow has a Help me choose guide in Settings and on first launch. It asks two questions, reads your graphics card and marks the model that suits you with Best for you. One button downloads it if needed and switches to it. This page gives the same rule in writing.
The two questions
- Which language do you mostly speak? Greek with English words mixed in, English only, or another language.
- What matters most to you? That it writes correctly, that it writes fast, or that your computer stays free, for example while you play.
The app also reads how much memory your graphics card has. That decides what fits, so the answer is personal to your machine.
What the guide picks
| Your situation | Best for you | Why |
|---|---|---|
| Greek or another language, card with 3 GB or more, you want it correct | Whisper large-v3 | The fewest mistakes, Greek above all. |
| Card with 3 GB or more, you want speed | Whisper large-v3-turbo q8 | Noticeably faster, with a few more mistakes in Greek. |
| English only, card with 3 GB or more | Whisper large-v3-turbo q8 (q5 for speed) | In English it comes close to the big model and writes much faster. |
| Card with 1.5 to 3 GB | Whisper large-v3-turbo q8 or q5 | It fits easily and writes well. |
| You want the computer free while you dictate | Whisper large-v3-turbo q5 | The lightest model that still writes well. |
| No graphics card the app can use | Whisper large-v3-turbo q5 on the processor | The fastest model on a processor that still writes well. |
The models in plain words
- large-v3: for Greek or another language besides English, on a card of 3 GB or more. You get the fewest mistakes. It is a little slower and takes about 2 GB on the card.
- large-v3-turbo q8: for speed with good accuracy, or for mostly English. Much faster than large-v3 and close to it in English. In Greek it makes more mistakes than large-v3.
- large-v3-turbo q5: for small cards, computers without a card, or people who play while they dictate. The lightest and fastest that still writes well, about 1 GB. A little less accurate than turbo q8.
- medium: for old computers where the others will not start. It loads almost anywhere and makes the most mistakes in Greek, so turbo q5 is usually the better pick.
Where the ranking comes from
On Greek, the published FLEURS figures give Whisper large-v3 10.9 wrong words out of 100 and large-v3-turbo 13.0. On Common Voice the same two score 13.7 and 20.6. On English the turbo model comes close to large-v3 and runs several times faster. The guide in the app was written on 23 September 2026 from those figures.
The card limits in the rule are 3000 MB for large-v3 and 1500 MB for turbo. Below 1500 MB, or with no usable card, the processor does the work. Other programs holding the card can still push a model out of it; the card memory guide shows how to see that.
More dictation guides
- Free Voice to Text on PC: Talk and It Types
- Fix Slow Offline Dictation: Whisper GPU Memory
- Windows Dictation Hotkey: Start, Stop and Cancel
- Fix Dictation Spelling with a Custom Dictionary
- Dictation Text Not Appearing? Fix Windows Paste
- Greek and English Dictation: Language Settings
- Free Greek and English Dictation in Word on Windows
- Voice Typing for ChatGPT and Claude on Windows
FU Flow 0.9.15 beta · 1.7 GB · Windows 10 / 11 · MIT