For years, Dragon NaturallySpeaking was the default answer to "what dictation software should I use?" It dominated the market through the 2000s and 2010s, building a loyal user base of doctors, lawyers, writers, and professionals who relied on voice dictation daily. But for Mac users, Dragon has been a dead end since 2018. Nuance discontinued Dragon for Mac, and the Windows version has received only incremental updates since Microsoft acquired Nuance in 2022. If you are on a Mac and searching for a Dragon replacement, here is how Steno compares.
The State of Dragon on Mac
Dragon Professional Individual for Mac was last updated in version 6.0.8. It requires Rosetta 2 to run on Apple Silicon Macs, which means it runs in an emulation layer rather than natively. Nuance has made it clear there will be no Apple Silicon native version, no macOS Sonoma or Sequoia updates, and no further development of any kind. The product is end-of-life.
For existing Dragon Mac users, this means living on borrowed time. Each macOS update risks breaking compatibility. Rosetta 2 itself has an uncertain future as Apple continues its transition away from Intel architecture. And even when Dragon still runs, it runs slowly. An emulated app on modern hardware delivers a worse experience than it did on the native Intel Macs it was designed for.
Dragon for Windows remains available at $699 for the Professional edition. It still receives updates, but it is a Windows-only product with no cross-platform plans. Mac users who need dictation software need to look elsewhere.
How Steno Differs from Dragon
Steno and Dragon represent fundamentally different approaches to dictation. Dragon was built in an era when speech recognition required extensive local processing, voice training sessions, and large acoustic models stored on disk. Steno was built for the current era, where cloud-based AI models like Whisper deliver superior accuracy without any training, and native Swift code means the app itself is under 2 MB.
Architecture
Dragon installs as a heavyweight desktop application. The Mac version was over 4 GB. It bundles its own speech recognition engine, acoustic models, language models, and a custom text editor. It manages its own user profiles, vocabulary files, and correction databases.
Steno is a 1.7 MB native macOS app that lives in your menu bar. It uses the Groq Whisper API for transcription, which means the speech recognition model is always up to date without any app updates. There are no user profiles to train, no vocabularies to import, and no correction databases to maintain. You install it, grant microphone and accessibility permissions, and start dictating.
Accuracy
Dragon's accuracy was impressive for its time. After training with your voice for 15 to 30 minutes, it could achieve word error rates of 5 to 8 percent for clear English dictation. It excelled at domain-specific vocabulary because users could add custom words and train the engine to recognize them.
Steno uses OpenAI's Whisper large-v3-turbo model, which achieves comparable or better accuracy out of the box with no training. Whisper was trained on 680,000 hours of multilingual audio data, giving it a breadth of vocabulary and accent handling that no single-user-trained model can match. For medical terminology, legal jargon, programming terms, and other specialized vocabulary, Steno also supports custom vocabulary lists and profession-specific modes that bias the transcription model toward domain terms.
Where Dragon required hours of correction and retraining to improve accuracy for new vocabulary, Steno's cloud model improves continuously through model updates that happen server-side, requiring no action from the user.
Interaction Model
Dragon uses a toggle dictation model. You click a button or press a hotkey to start dictation, speak for as long as you want, and click or press again to stop. The system has to determine when you are pausing versus when you are finished, which leads to endpoint detection errors and accidental captures.
Steno uses a hold-to-speak model. You hold a hotkey, speak, and release when you are done. The microphone is only active while your finger is on the key. This eliminates accidental captures, endpoint ambiguity, and the cognitive overhead of managing a hidden dictation state. It is the same push-to-talk interaction used in aviation radio, gaming voice chat, and walkie-talkies, chosen because it gives the speaker absolute control over when their voice is captured.
Smart Rewrite vs. Raw Transcription
Dragon outputs raw transcription. What you say is what you get, with some basic formatting rules applied. If you say "um" or "you know" or stumble over a word, Dragon transcribes it.
Steno includes Smart Rewrite, an AI post-processing layer that polishes the raw transcription based on context. It detects the app you are typing into and adjusts tone accordingly. Dictating in Slack produces casual, lowercase text. Dictating in Mail produces professional, properly formatted prose. Dictating in VS Code preserves technical terms, variable names, and code formatting. This context-aware polishing happens automatically, so the text that appears at your cursor is ready to send without manual editing.
Language Support
Dragon supports English with specialized vocabulary packs for medical and legal domains. Additional language packs are available for major European languages, each requiring separate installation and configuration.
Steno supports over 90 languages out of the box through Whisper's multilingual model. Language detection is automatic. You can switch between English and Hindi mid-sentence, and Steno will handle both correctly. There is nothing to install, configure, or switch. You simply speak in whatever language feels natural.
Feature Comparison
- Platform: Dragon is Windows only (Mac discontinued). Steno is native macOS, Apple Silicon optimized.
- App size: Dragon is 4+ GB. Steno is 1.7 MB.
- Setup time: Dragon requires 15 to 30 minutes of voice training. Steno requires zero training.
- Price: Dragon Professional is $699 one-time. Steno is free with a Pro tier at $4.99 per month.
- Accuracy model: Dragon uses a locally trained acoustic model. Steno uses Whisper large-v3-turbo via Groq.
- Languages: Dragon supports English plus select European languages. Steno supports 90+ languages with automatic detection.
- Post-processing: Dragon outputs raw transcription. Steno includes Smart Rewrite with context-aware tone adjustment.
- Voice commands: Dragon has extensive voice command vocabulary. Steno supports punctuation commands, text snippets, and profession-specific commands.
- Offline mode: Dragon works fully offline. Steno supports offline mode via a local Whisper model, with cloud as the default for maximum accuracy.
- Updates: Dragon for Mac receives no updates. Steno receives regular updates via Sparkle auto-update.
Who Should Choose Which
If you are on Windows and already have a trained Dragon profile with years of custom vocabulary, Dragon still works and there is no urgent reason to switch. Your investment in training the model has real value, and Dragon's voice command system remains more extensive than any competitor.
If you are on a Mac, Dragon is no longer a viable option. The discontinued Mac version is running on borrowed time, and every macOS update increases the risk that it stops working entirely. Steno is the natural replacement: a native macOS app built in Swift, optimized for Apple Silicon, with modern AI-powered accuracy that matches or exceeds Dragon's best results without any training period.
If you are a new user on any platform evaluating dictation software for the first time, the calculus is straightforward. Steno offers better accuracy out of the box, costs a fraction of Dragon's price, requires no training, supports 90+ languages, and includes Smart Rewrite for context-aware text polishing. You can download it for free from stenofast.com and be dictating within 30 seconds.
Dragon defined dictation for two decades. But the technology has moved on. AI-powered models like Whisper deliver better accuracy with zero training, and native apps like Steno deliver it in a package that is 2,000 times smaller.