Works in every app
Chat boxes, editors, terminals, browsers, mail. If you can paste into it, you can dictate into it. One hotkey, system-wide.
- AI chat
- Code editor
- Terminal
- Browser
- Notes
- Docs
Windows beta. Free 15 min a month, no card.
Hold a key. Speak. It’s typed.
System-wide dictation for Windows. Speak any language, release the key, and your words land in the app you’re using.
Animated illustration: holding Right Ctrl opens a small Listening pill with a live level meter. Releasing the key shows Transcribing, then Copied, and a Greek sentence lands in a chat box with the technical terms endpoint, Hono, validate, JWT and 401 kept in Latin script.
Three moves. No window switching.
Sign in with an email link or code when you’re ready. No password.
Step 1: Hold your hotkey.
Right Ctrl by default. Press and hold it anywhere in Windows. You can change the key, or tap it once for hands-free mode.
Step 2: Speak.
Any language. Lock one, like Greek, or let it detect the language for you. Mix in English technical terms the way you normally do.
Step 3: Release.
Your text is ready in about half a second, usually 0.3-0.6 s. It’s on your clipboard. Turn on auto-paste and it goes straight into the app you’re in.
Built for people who talk to computers all day
Hands-free mode
Tap the hotkey once to start, tap again to stop. For long thoughts, or when holding a key isn’t practical.
Speak any language, or lock yours
Auto-detect handles mixed days. Lock Greek, or another language, and it stops guessing.
Technical terms stay in Latin script
Say “refactor the useEffect hook” in Greek and get useEffect back, not a phonetic guess. Code identifiers, product names and acronyms stay as you’d type them.
Κάνε refactor το useEffect hook στο Dashboard.tsx και πρόσθεσε cleanup.
Auto-copy, plus opt-in auto-paste
By default your text lands on the clipboard. Prefer it typed straight into the active app? Turn on auto-paste in settings.
- Auto-copy
- On
- Auto-paste
- Off
Private by design
Your audio is processed transiently to produce your text. Our speech-to-text provider is configured for zero data retention and never trains on it. We keep usage metadata only, so we can count your minutes.
- Kept
- Seconds used, mode, detected language
- Never stored by us
- Audio, transcripts
Coming soon
Dragon mode: say it in your language. Paste it in Simplified Chinese.
Speak in any language. Dragon mode will translate what you said into Simplified Chinese and paste that instead. It’s built for one thing: spending fewer tokens when you talk to an AI.
Who this helps
If you write to AI tools in Greek, or another language that tokenizes poorly, your prompts can cost more than they need to. In our test the same messages took 1.9x as many tokens in Greek as in English on the GPT-4o tokenizer, and 4.2x on GPT-4’s. Simplified Chinese is far denser.
Who this doesn’t help
If you already write in English, Chinese will not save you tokens. In our test it took 13% more tokens than English on the GPT-4o tokenizer and 47% more on GPT-4’s. We’re telling you now so you don’t find out later.
How we compare
Savings are always measured against the language you actually write in. The only English number on this page is the one above, where Chinese costs more. All figures come from OpenAI’s open tokenizers; other models may differ.
A trade-off to know about
Prompts in Chinese can change how well a model follows your task. Dragon mode will be a toggle: on for the prompts where cost matters, off where precision matters most.
How many tokens would you save?
Pick your language and how much you dictate. We use measured token counts for the same messages in your language and in Simplified Chinese, on the same tokenizer. Method, data and script
- In Greek
- 4,450tokens / day
- In Simplified Chinese
- 2,700tokens / day
−40%Chinese vs. Greek
How we measured
We wrote 12 realistic messages to an AI assistant in English (23 to 33 words each: 4 coding requests with English identifiers, 2 writing help, 2 analysis, 2 everyday, 1 CI/devops, 1 personal finance explanation), translated each one into every language and into Simplified Chinese, and counted the tokens of every message separately with tiktoken 0.14.0 on 2 October 2026.
Change = (tokens in Chinese − tokens in your language) / tokens in your language, over the whole set. A negative number means Chinese uses fewer tokens.
- GPT-4o / GPT-5: OpenAI o200k_base (GPT-4o, GPT-4.1, GPT-5 family)
- GPT-4: OpenAI cl100k_base (GPT-4, GPT-3.5-turbo)
Code identifiers, file paths and product names stay in Latin script in every language (as Dragon mode will keep them). Each language uses its own natural conventions for numerals and common technical loanwords.
Limits
- Small corpus (12 messages per language); treat results as indicative, not as a benchmark.
- Translations were written by hand for this measurement; a different translator or Dragon mode's live model output may be longer or shorter, especially for zh-Hans.
- Results depend on the domain mix: messages full of code identifiers save less, because identifiers stay in Latin script and cost the same in every language.
- Only OpenAI's open tokenizers are measured; other models (Claude, Gemini, Llama, Qwen) use different tokenizers and will give different numbers.
- Only input tokens are measured. Model replies are not part of this comparison.
- For English, Chinese costs MORE tokens; Dragon mode is not a token saver for English speakers.
- Claude (Anthropic) uses its own tokenizer and is not part of these numbers yet.
Chinese against English: 13% more tokens on the GPT-4o tokenizer and 47% more on GPT-4’s. That is why we never compare against English.
Full methodology, per-language results and the measurement script
Notify me about Dragon mode
Same email list, separate tag. One message when it’s live.
Start free. Go Pro when you dictate every day.
Prices in US dollars and euros. Euro prices include VAT. No card for Free.
Free
Try it on real work.
$0 / €0forever
No card required
15 min / month
- 15 minutes of dictation a month
- No card required
- Works in every app
- Any language, or lock yours
- Auto-copy, opt-in auto-paste, hands-free mode
- Audio processed transiently, never stored by us
Pro
For people who dictate every day.
$16 / €16per month
$16 + applicable tax,€16 incl. VAT
Or $144 / €144 a year, save 25%
$144 / €144per year
$144 + applicable tax,€144 incl. VAT
$12 / €12 a month, billed yearly. Save 25%
4,000 min / month
- 4,000 minutes of dictation a month, over 2 hours a day
- Everything in Free
- Monthly or yearly billing
- Cancel anytime
Coming
- Custom dictionary
- Priority support
We let you know at 80%. Minutes reset monthly and don't roll over.
Join the Windows betaDragon mode is coming soon. When it launches it will be included in every plan, Free too.
- Minutes are audio minutes. We count the length of your recorded speech, not the time the app is open. They reset monthly and don’t roll over.
- Pro is 4,000 minutes a month, over 2 hours a day of speech. We tell you at 80%, and dictation pauses at the limit until your minutes reset.
- Taxes. Euro prices include VAT and are the same in every EEA country. US dollar prices are before any applicable tax, which Paddle adds at checkout. Paddle, our Merchant of Record, shows the final price at checkout.
- Changed your mind? Full refund within 14 days of your first purchase, and you can withdraw from the contract as the law allows.
- Paid plans open with the beta. Join the waitlist and we’ll tell you.
Questions, answered
Which apps does it work in?
Almost any app where you can type: browsers, chat apps, editors, terminals, email, notes. By default Dragon Wispr copies your text to the clipboard, so you can paste anywhere. Turn on auto-paste and it pastes into the active window for you. A few protected windows, such as apps running as administrator, can block automatic pasting. Your text is still on the clipboard in that case.
Is my audio stored?
Not by us. Your audio is processed transiently to produce your text. Our speech-to-text provider is configured for zero data retention and never trains on it. We never store audio or transcripts. We keep usage metadata only: how many seconds you dictated, the mode, and the detected language, so we can count your monthly minutes.
Which AI models process my speech?
Speech-to-text runs on Whisper large-v3 turbo, an open-weights model, hosted by Groq. When Dragon mode launches, translation will use Qwen3-32B, an open-weights model, hosted by DeepInfra. Neither provider may train on your dictations, and neither do we. Like any AI, these models can make mistakes, especially with names, numbers and technical terms, so read the text before you send it. If we change a provider or model, we update our privacy policy first.
Which languages does it understand?
Dragon Wispr detects the language you speak. If you always speak one, like Greek, you can lock it and skip detection. We will publish the full supported list at launch.
Will it mangle my English technical terms?
That is the case we design for. Technical terms, code identifiers, product names and acronyms stay in Latin script inside your sentence, so a Greek sentence with endpoint, JWT and useEffect comes out the way you would type it.
How fast is it?
Typically 0.3-0.6 s from releasing the key to text, measured in our own testing. Your connection and the length of what you said move that up or down.
Does Chinese really save tokens?
Compared with your own language, often yes. In our test the same messages took 1.9x as many tokens in Greek as in English on the GPT-4o tokenizer, and Simplified Chinese used 40% fewer tokens than the Greek. Compared with English, no: Chinese took 13% more tokens than English on the same tokenizer. If you write in English, Dragon mode will not help you. Every number on this site compares Chinese with your own language, and we publish the method, the test messages and the script. Dragon mode is coming soon.
Can Chinese prompts make AI answers worse?
For some tasks, yes. Published coding benchmarks show models solving somewhat fewer tasks when the prompt is in Chinese instead of English, and how Chinese compares with your own language depends on the model and the task. Dragon mode will be a toggle: use it where token cost matters, and keep it off where precision matters most.
What happens when I hit my limit?
On Free, dictation pauses when you have used your 15 minutes, until they reset at the start of your next monthly period or until you upgrade to Pro. Pro includes 4,000 minutes a month (over 2 hours a day); when you reach it, dictation pauses until your minutes reset at the start of your next monthly period. On both plans the app tells you when you have used 80%.
Do I need a card for the Free plan?
No. Free is 15 minutes a month with no card. You sign in with an email link or code, no password needed.
How much is Pro, and in which currency?
Pro is $16 / €16 a month, or $144 / €144 a year, which works out to $12 / €12 a month and saves 25%. Euro prices include VAT and are the same in every EEA country. US dollar prices are before any applicable tax, which Paddle adds at checkout. You can cancel anytime.
Is VAT or sales tax included?
Euro prices include VAT, and the euro price is the same in every EEA country. US dollar prices are shown before tax: Paddle, our Merchant of Record, adds any sales tax or VAT that applies where you live. Either way you see the final price at checkout before you pay, and Paddle issues your invoice.
Is there a macOS version?
Not yet. Windows comes first and macOS is planned. Join the waitlist and we will let you know.
Is there a download yet?
Not yet. Join the waitlist and we will email you when the Windows beta opens.
Refunds?
Yes: a full refund within 14 days of your first purchase of a paid plan, no questions asked. Payments are processed by Paddle, our Merchant of Record. As an EU consumer you can also withdraw from the contract within 14 days. You can cancel anytime to stop future charges.
Can I download or delete my data?
Yes. In the desktop app's account settings, Download my data gives you a file with everything we hold about you, and Delete account removes your account and its data straight away. You can also email privacy@dragonwispr.com.
Hold a key. Speak. It’s typed.
Join the Windows beta. Free, 15 minutes a month, no card.