ElevenLabs Scribe
Lowest errors across languages on the 2026 leaderboards, with speaker labels.
Free, then $6/mo
Best speech to text (transcription)
Short answer: ElevenLabs Scribe. For the most accurate transcript, use ElevenLabs Scribe v2.
Which AI turns audio into text most accurately?
Buzz
Free app, no minute limits, runs offline on Windows, Mac and Linux.
FreeEasyOpen
Whisper large-v3 turbo
Runs offline in free apps on any computer, in 99 languages.
FreeEasyOpen
MacWhisper
Pay once, transcribe unlimited hours privately on your Mac.
FreeEasy
Mistral launched Voxtral Transcribe 3 in October 2026 and Google released Gemini 3.5 Transcribe in August.
All 11 side by side.
Price, whether it runs on your own computer, and the main catch.
| # | Tool | Best for | Price | On your PC | Main downside |
|---|---|---|---|---|---|
| 01 | ElevenLabs ScribeElevenLabsBest overall | Interviews, podcasts, multilingual audio | Free, then $6/mo | No | Free minutes run out in one short meeting |
| 02 | Gemini 3.5 TranscribeGoogle | Dictation and clean written transcripts | Free | No | Free API use may be used to improve Google products |
| 03 | GPT TranscribeOpenAI | Apps and scripts that transcribe audio | Paid | No | Developer API only, no simple app for files |
| 04 | MacWhisperGood Snooze (Jordi Bruin) | Mac users with lots of recordings | Free | Easy | Mac only |
| 05 | AssemblyAIAssemblyAI | Developers transcribing many hours | Free | No | Made for developers, not a point-and-click app |
| 06 | Voxtral Transcribe 3Mistral AI | Low-cost accurate transcription by API | Paid | Smaller version | Version 3 weights are not downloadable, unlike earlier Voxtral |
| 07 | Whisper large-v3 turboOpenAI | Free offline transcription, 99 languages | Free | Easy | No longer the most accurate, newer open models beat it |
| 08 | NVIDIA Parakeet TDT v3NVIDIA | Fast offline transcription in European languages | Free | Some setup | Only 25 European languages, no Chinese, Japanese or Hindi |
| 09 | BuzzChidi Williams (open source) | Free private transcription and subtitles | Free | Easy | Windows installer is unsigned, so you get a warning |
| 10 | DeepgramDeepgram | Real-time captions and voice agents | Free | No | Error rate now well behind the leaders |
| 11 | RevRev.com | Legal, research and when accuracy matters | Free, then $29.99/mo | No | AI accuracy behind newer models on public tests |
Nothing matches. Try another filter.
Pros and cons of all 11
-
ElevenLabs Scribe
ElevenLabs · Scribe v2
The most accurate transcription across many languages right now.
Best overallFree plan- Price
- Free plan, then $6/moFree plan includes about 12 minutes of transcription a month.
- On your PC
- No, online only
- Top score on the multilingual Open ASR leaderboard
- Labels who is speaking and marks laughter, music
- Free minutes run out in one short meeting
- Credits are shared with voices, so heavy use gets pricey
-
Gemini 3.5 Transcribe
Google · Gemini 3.5 Transcribe
Google's newest transcription model, cleans up ums and restarts.
Free- Price
- FreeFree tier in Google AI Studio, with usage limits.
- On your PC
- No, online only
- Very low error rate, among the top ten
- Removes filler words and self-corrections automatically
- Free API use may be used to improve Google products
- Consumer apps so far only Mac Gemini app and Gboard
-
GPT Transcribe
OpenAI · gpt-transcribe
OpenAI's newest transcription model, the successor to Whisper.
- Price
- Paid
- On your PC
- No, online only
- Third best on the multilingual Open ASR leaderboard
- Cheaper per minute than the old Whisper API
- Developer API only, no simple app for files
- Less accurate than Scribe v2 and Voxtral 3 on tests
-
MacWhisper
Good Snooze (Jordi Bruin)
The easiest way to transcribe files privately on a Mac.
Best valueFreeRuns on your PC- Price
- FreeFree version with smaller Whisper models, runs offline.
- On your PC
- EasyApple Silicon Mac with 8 GB memory
- Drag a file in, get text, nothing leaves your Mac
- Runs both Whisper and the faster Parakeet models
- Mac only
- Best models and speaker labels need Pro
-
AssemblyAI
AssemblyAI · Universal-3.5 Pro
Accurate, cheap transcription API with speaker labels and summaries.
Free- Price
- Free$50 free credit on signup, no card needed.
- On your PC
- No, online only
- Very low price for this accuracy
- Fourth on the multilingual Open ASR leaderboard
- Made for developers, not a point-and-click app
- Streaming is billed while the connection is open
-
Voxtral Transcribe 3
Mistral AI · Voxtral Small Transcribe 3 / Mini Transcribe 3
Mistral's new transcription models, among the most accurate and cheap.
Smaller version local- Price
- Paid
- On your PC
- TechnicalNVIDIA GPU with 16 GB memory for the open 4B model
- Third lowest error rate on Artificial Analysis
- Among the cheapest top models per minute
- Version 3 weights are not downloadable, unlike earlier Voxtral
- Developer API only, no consumer transcription app
-
Whisper large-v3 turbo
OpenAI · large-v3-turbo
The famous free open model behind most transcription apps.
Best on your computerFreeOpen sourceRuns on your PC- Price
- Free, open sourceFree to download and run on your own computer.
- On your PC
- EasyAny recent laptop with 8 GB RAM; faster with a GPU
- Free, offline and private on your own computer
- Works in 99 languages
- No longer the most accurate, newer open models beat it
- Much slower than Parakeet on the same computer
-
NVIDIA Parakeet TDT v3
NVIDIA · parakeet-tdt-0.6b-v3
Small open model, more accurate and far faster than Whisper.
FreeOpen sourceRuns on your PC- Price
- Free, open sourceFree to download and use, commercial use allowed.
- On your PC
- Some setupAny recent laptop with 4 GB RAM; fastest with NVIDIA GPU
- Beats Whisper turbo on the English leaderboard
- Very fast, transcribes an hour in seconds on a GPU
- Only 25 European languages, no Chinese, Japanese or Hindi
- Portuguese training is European, not Brazilian
-
Buzz
Chidi Williams (open source) · 1.4.5
Free desktop app that runs Whisper offline on Windows, Mac, Linux.
Best freeFreeOpen sourceRuns on your PC- Price
- Free, open sourceCompletely free, no limits, everything stays on your computer.
- On your PC
- EasyAny laptop with 8 GB RAM; faster with a GPU
- Free with no minute limits
- Works on Windows, Mac and Linux
- Windows installer is unsigned, so you get a warning
- Plainer and slower than MacWhisper
-
- Price
- Free$200 free credit that does not expire, no card needed.
- On your PC
- No, online only
- Biggest free credit of the big APIs
- Very fast live transcription for voice apps
- Error rate now well behind the leaders
- Developer tool, not an app for normal users
-
- Price
- Free plan, then $29.99/mo45 minutes of AI transcription a month.
- On your PC
- No, online only
- Real people can check and fix the transcript
- Simple app for recording and uploading
- AI accuracy behind newer models on public tests
- Expensive monthly plans compared with others
5 of these run on your own computer.
Private and offline. Check what your machine can handle.
| Tool | Setup | You need |
|---|---|---|
| MacWhisper | Easy | Apple Silicon Mac with 8 GB memory |
| Voxtral Transcribe 3 | Smaller version | NVIDIA GPU with 16 GB memory for the open 4B model |
| Whisper large-v3 turbo | Easy | Any recent laptop with 8 GB RAM; faster with a GPU |
| NVIDIA Parakeet TDT v3 | Some setup | Any recent laptop with 4 GB RAM; fastest with NVIDIA GPU |
| Buzz | Easy | Any laptop with 8 GB RAM; faster with a GPU |
Quick answers.
What is the best free speech to text?
Buzz (Windows, Mac, Linux) and the free MacWhisper run Whisper on your own computer with no minute limits. Online, Rev gives 45 free minutes a month and ElevenLabs about 12.
Can I transcribe audio offline?
Yes. Whisper and NVIDIA Parakeet run fully offline in apps like MacWhisper and Buzz. A normal laptop with 8 GB of memory is enough; a graphics card just makes it faster.
Is Whisper still the best?
No. Whisper is still the most used free model, but on the Open ASR leaderboard newer models like Parakeet, Qwen3-ASR and paid services like ElevenLabs Scribe make fewer mistakes.
ElevenLabs Scribe vs OpenAI GPT Transcribe?
Scribe v2 has fewer errors on both public leaderboards and labels speakers. GPT Transcribe is a cheap developer API at $0.0045 a minute with no consumer app.
What is word error rate (WER)?
The share of words the AI gets wrong. A WER of 3 means about 3 mistakes per 100 words. Lower is better.
Rankings we read: Hugging Face Open ASR Leaderboard, Artificial Analysis speech to text. Prices from each official pricing page, checked 2026-11-05.
People also compare.
Need it built for your company instead?
Vomini builds AI systems that run inside your business, with your data and your rules. You see it working before you pay.