Came out right
- Flula
- 81%
- Wispr Flow
- 67%
The share of dictations that passed every check: the right words, nothing extra, laid out as they should be.
Flula vs Wispr Flow.
Both turn speech into clean text. On the same dictations, Flula got 81% right and Wispr Flow 67%, just as fast.
Tested by us on 4–6 November 2026; how we measured is at the bottom. Features checked against Wispr Flow’s public pages on 4 October 2026.
Our test, November 2026
Same dictations,
Same dictations,
both apps.
Scripts in 19 languages, and the same recordings played into each app. How we measured
Words wrong
- Flula
- 4.4%
- Wispr Flow
- 9.3%
The average share of words that differ from the text we expected. Flula gets about half as many wrong.
Stop to text
- Flula
- 0.99 s
- Wispr Flow
- 1.01 s
The average wait for your text after you stop talking, over every dictation in all 19 languages, not just English. Wispr Flow was quicker in English; across the other languages, Flula was.
| Kind of dictation | Flula | Wispr Flow |
|---|---|---|
| Spoken corrections, like “Tuesday, no, Wednesday” | 85% | 52% |
| Fillers and repeats taken out | 86% | 52% |
| Names of people and places | 60% | 32% |
| Email addresses, phone numbers and addresses | 73% | 52% |
| Numbers, dates, times and prices | 90% | 78% |
| Switching languages mid-sentence | 80% | 69% |
| Messages, emails and notes | 81% | 66% |
| Instructions meant as text, like “Translate this into French: …” | 89% | 78% |
| A few words at a time | 91% | 87% |
| Pauses while you think | 68% | 64% |
| Plain speech | 97% | 94% |
| Prompts for AI chat apps | 90% | 95% |
| Spoken formatting: new lines and lists | 64% | 72% |
| Language | Flula | Wispr Flow |
|---|---|---|
| Arabic | 75% | 78% |
| Chinese | 89% | 82% |
| Dutch | 92% | 53% |
| English | 87% | 82% |
| Estonian | 71% | 56% |
| French | 72% | 78% |
| German | 89% | 88% |
| Hindi | 53% | 53% |
| Indonesian | 75% | 47% |
| Italian | 89% | 78% |
| Japanese | 83% | 47% |
| Korean | 78% | 61% |
| Polish | 92% | 61% |
| Portuguese | 75% | 58% |
| Russian | 92% | 75% |
| Spanish | 75% | 69% |
| Turkish | 83% | 53% |
| Ukrainian | 83% | 64% |
| Vietnamese | 83% | 58% |
| Two languages in one dictation | 77% | 61% |
marks the higher share of dictations right. Forms, punctuation, technical terms and sound-alike words came out the same for both.
Side by side
Feature by feature.
| What you get | Flula | Wispr Flow |
|---|---|---|
| Dictations right in our test | 81% of dictations right, with about half as many words wrong | 67% |
| Speed in our test | Your text about a second after you stop: 0.99 s on average across all 19 languages, and quicker outside English | About the same: 1.01 s on average, and quicker in English: 0.43 s against 0.64 s |
| Switching languages mid-sentence | Built for it: pick the languages you speak once, then mix them in one sentence. In our test, 77% of mixed-language dictations came out right | Detects one language per dictation, not per word; its docs say mixed-language speech is less reliable. In our test, 61% came out right |
| Settings per app | Yes, on every plan: a tone for each kind of app (messages, work chat, email, the rest) and, on Mac, a language for any app | Yes, for style: apps are grouped into categories such as messages and email. English only; the language doesn’t change per app |
| Translation | Opt-in, into any of 101 languages. Off by default | No output-language setting. Translates only through a custom instruction, or on desktop a translate-to-English action |
| Clean-up | Adds punctuation, removes fillers, applies spoken corrections. In our test it got fillers right 86% of the time and corrections 85% | Adds punctuation, removes fillers and applies spoken corrections, with three clean-up levels. In our test, 52% and 52% |
| Where your speech goes | Processed online, never saved | Processed in the cloud. On Free and Pro, sharing audio and transcripts for model training is on by default; you can turn it off |
| Languages | 101 | 100+ |
| Platforms | Mac. iPhone soon | Mac, Windows, iPhone, Android |
| Free plan | 20 minutes a month (about 3,000 words) | 2,000 words a week on desktop, fewer on mobile |
marks the stronger option on a row, based on each product’s own published information, or on our test where the row gives its numbers. Unmarked rows are close, or depend on what you need.
Heracles once stood where the road forked. Either road can be the right one.
Choose Flula if…
- You want less to fix: in our test Flula got 81% of dictations right, Wispr Flow 67%
- You switch languages mid-sentence: Flula keeps each word in the language you said it
- You want an app to write in its own language on your Mac, not just in its own style
- You want a setting that writes in any of 101 languages, only when you choose it
- You don’t want your audio kept or used for training: Flula never saves it
Choose Wispr Flow if…
- You want unlimited dictation on a paid plan
- You want a bigger free allowance: 2,000 words a week on desktop
- You need Windows or Android
- You dictate a lot of lists or prompts for AI chat apps, which Wispr Flow did a little better in our test
Where Flula is different
Made for people who
Made for people who
switch languages mid-sentence.
Mix languages mid-sentence
Pick the languages you speak once. Flula keeps every word in the language you said it.
A tone for each kind of app
Casual in Slack, polished in Mail, translated in the apps you choose. On a Mac, Flula switches for you.
Translation only when you ask
Output stays as spoken by default. Choose a language and Flula writes in it.
How we measured
Our test,
Our test,
and what it can’t tell you.
What we did
- The test is a set of dictations in 19 languages, many of them mixing languages. Each is a script of what someone says, with the ums, the “no, wait” and the spoken “new line”, plus the text a careful editor would type from it.
- They were written after Flula was tuned, and nobody tuning Flula reads them or the results, so Flula can’t have been fitted to them.
- Each app heard the same recordings, one app at a time, set to the languages being spoken.
- We used the current version of each app in early November 2026, with its usual clean-up settings.
- A dictation counts as right when it passes every check: the words, names and numbers are there, fillers and taken-back words are gone, and new lines and lists land where they should. Speed is the average wait from letting go of the key until the text is in, over every dictation in every language.
What to keep in mind
- It’s our test: our scripts, our checks, run by us. Your voice, accent and microphone will give different numbers.
- A few dictations come with their own word list, and Wispr Flow’s dictionary didn’t have those words. Leave them out and it’s 82% against 68%.
- A few recordings longer than 49 seconds didn’t run on Wispr Flow, so they aren’t counted for either app.
- A few languages ran with Flula features we haven’t released to everyone yet. Leave them out and Flula got 79% right, Wispr Flow 68%.
- Both apps change often. This is one run of each, in November 2026.
Sources
Wispr Flow’s pricing page, help center and data-controls page, checked 4 October 2026. Our own test of both apps, run 4–6 November 2026; we keep the recordings, settings and every result so the numbers can be checked.
Wispr Flow is a trademark of its owner. Prices and features change; check their site for current details. Flula is not affiliated with Wispr Flow.
Try both. Keep one.
Flula’s free plan: 20 minutes a month, every language, no card.