By No Code MBA · Facts checked
- Provider
- Mistral
- Announcement date
- Status at last check
- Generally available
- Model family
- Voxtral
- Preceding model
- Voxtral Mini Transcribe 2 and Voxtral TTS
- Capabilities
- Audio, Text
What changed
Voxtral 3 is a speech family of four models: Voxtral Small Transcribe 3 and Voxtral Mini Transcribe 3 (4B parameters) for batch transcription with diarization in 50+ languages, Voxtral Mini Transcribe Realtime 3 with sub-500ms latency and end-of-turn detection, and Voxtral Mini TTS 3 with voice cloning in 35+ languages. A voice agent API combines listening, reasoning, tool calling and speech with turns under one second. Mistral reports Small Transcribe 3 beats Whisper large-v3, Scribe v2 and Gemini 3.5 Transcribe on FLEURS and diarization.
Access and availability
Available through Mistral's API with model names such as voxtral-small-transcribe-3-0, voxtral-mini-transcribe-realtime-3-0 and voxtral-mini-tts-3-0. Mistral's docs list all four as Premier models, and on-premises or private-cloud self-deployment requires contacting Mistral's sales team.
Official access documentationPricing
Small Transcribe 3: $0.004 per minute; Mini Transcribe 3: $0.0035 per minute (batch); Mini Transcribe Realtime 3: $0.008 per minute; Mini TTS 3: $13 per million characters; voice agent API: $0.04 per minute with Mistral Small 4 or $0.08 per minute with a larger model
Official pricingWhat you can build
Evaluate it on a short, representative voice workflow that combines transcription, a tool call, and a spoken response, then review accuracy, latency, and handoff behavior on your own samples.
Limitations
Voxtral 3 models are not listed as open weights, unlike earlier Voxtral releases. Benchmarks are Mistral's own (TTS quality was judged by an LLM), most model sizes are undisclosed, and preset voices cover only 18 languages.