Affiliate disclosure: Top10AI has no affiliate relationship with Mistral AI or ElevenLabs. The link in the “alternative to try” section is an affiliate link, which means we may earn a commission at no extra cost to you.
Mistral released Voxtral 3 on October 20, 2026: a family of AI speech-to-text and text-to-speech models plus a Voice Agents SDK. That puts it head to head with ElevenLabs, whose Scribe v2 transcription and Eleven v4 voice models are the default choice for many developers. Here is how the two compare, based only on each company’s official documentation. We have not run our own benchmarks.
Voxtral 3 vs ElevenLabs at a glance
| Mistral Voxtral 3 | ElevenLabs | |
|---|---|---|
| Transcription models | Voxtral Small Transcribe 3, Voxtral Mini Transcribe 3, Voxtral Mini Transcribe Realtime 3 | Scribe v2, Scribe v2 Realtime, Scribe v2 Medical |
| Transcription languages | 50+ languages | 90+ languages |
| Text-to-speech | Voxtral Mini TTS 3: 35+ languages and 50+ dialects | Eleven v4 and Eleven v4 Turbo: 90+ languages; Flash v2.5 for ultra-low latency |
| Real-time | Realtime transcription model; Voice Agents SDK for live phone and web conversations | Scribe v2 Realtime (about 150ms per ElevenLabs); Eleven v4 Turbo (about 100ms) |
| Diarization | Yes, with timestamps | Yes, up to 32 speakers |
| Voice cloning and design | Zero-shot cloning; “Pro Voice” adaptation from about 30 minutes of audio | Eleven TTV v3 voice design model (text to voice) |
| Deployment | EU-hosted API or self-deployment on your own infrastructure | ElevenLabs cloud API; HIPAA BAA available via sales |
| Published transcription price | Voxtral Small Transcribe 3: $0.004 per minute of input | See ElevenLabs pricing |
What’s new in Voxtral 3
Mistral frames Voxtral 3 around voice agents that can listen, reason, speak and take actions in one real-time pipeline. The release includes four models, all in Public Preview at launch:
- Voxtral Small Transcribe 3: Mistral’s flagship transcription and diarization model, priced at $0.004 per minute of input.
- Voxtral Mini Transcribe 3: a 4B-parameter model for offline transcription and diarization, which Mistral says costs $0.0035 per minute.
- Voxtral Mini Transcribe Realtime 3: real-time transcription.
- Voxtral Mini TTS 3: text-to-speech with zero-shot voice cloning, streaming and delivery that adapts emotion, pace and emphasis to the conversation.
Mistral also launched Voice Agents, a managed real-time pipeline (microphone capture, speech-to-text, LLM, text-to-speech and playback) with telephony, browser and custom-audio interfaces.
Accuracy: what Mistral claims
Mistral says Voxtral Small Transcribe 3 outperforms Whisper large-v3, GPT-Transcribe, ElevenLabs Scribe v2 and Gemini 3.5 Transcribe on transcription. In Mistral’s published chart, the macro-average word error rate over 45 FLEURS languages is 4.1% for Voxtral Small Transcribe 3 versus 4.9% for Scribe v2. Mistral’s diarization chart also shows Voxtral ahead of Scribe v2.
Keep in mind these are vendor-published benchmarks. Results on your own audio (accents, noise, domain terms) can differ. Run a short test set before switching providers.
Where ElevenLabs still leads
- Language coverage: ElevenLabs lists 90+ languages for Scribe v2 and for Eleven v4 speech, compared with Voxtral’s 50+ for transcription and 35+ for speech.
- Transcription features: ElevenLabs documents keyterm prompting (up to 1,000 terms), detection of 65 entity types, diarization of up to 32 speakers and a medical model that it says makes 35% fewer errors on clinical audio than Scribe v2.
Which should you choose?
- Choose Voxtral 3 if you need EU hosting or self-deployment for data sovereignty, want an integrated voice-agent pipeline from one vendor, or are price-sensitive on bulk transcription. Note that all four Voxtral 3 models launched in Public Preview.
- Choose ElevenLabs if you need the widest language coverage, expressive voices in 90+ languages, or medical transcription.
An alternative to try
If you don’t want to build on an API and just need transcripts and written content from recordings, Castmagic is an AI tool that turns long-form audio into transcripts, notes, summaries, quotes and social posts. It is not a Voxtral or ElevenLabs product, and we have not compared their accuracy.
See also our guides to AI note takers and Otter AI transcription.
FAQ
When was Voxtral 3 released?
Mistral released the Voxtral 3 family on October 20, 2026.
Is Voxtral 3 more accurate than ElevenLabs Scribe v2?
Mistral’s own benchmarks say Voxtral Small Transcribe 3 has a lower word error rate than Scribe v2. These are vendor results, not independent tests.
Can I run Voxtral 3 on my own servers?
Yes. Mistral says Voxtral 3 is available through its EU-hosted API or for self-deployment on your own infrastructure.
How many languages does each support?
Voxtral 3 transcribes 50+ languages and speaks 35+ languages and 50+ dialects. ElevenLabs lists 90+ languages for Scribe v2 and Eleven v4.
Sources
- Mistral AI, Voxtral 3 (announcement, October 20, 2026; checked October 23, 2026).
- Mistral Docs, Changelog, October 20 entry (checked October 23, 2026).
- Mistral Docs, Voxtral Small Transcribe 3 model card (checked October 23, 2026).
- ElevenLabs Docs, Transcription (Speech to Text) (checked October 23, 2026).
- ElevenLabs Docs, Models (checked October 23, 2026).
Related Reading
Find the Right AI Tool for You
Not sure which tool fits your workflow? Take our 30-second finder quiz and get a recommendation.
