RELEASE · MODELS · #2262
Mistral releases Voxtral 3 voice-agent SDK and multilingual speech models
Mistral announced Voxtral 3, a family of speech-to-text and text-to-speech models plus a real-time voice agent SDK for multi-step conversations in 50+ languages; components include Voxtral Small Transcribe 3, Voxtral Mini TTS 3, and Pro Voice. The offering is available via Mistral’s EU-hosted API or for self-deployment, and Mistral says the Transcribe models achieve state-of-the-art diarization and multilingual transcription, claiming outperformance of Whisper, Scribe v2 and Gemini 3.5 Transcribe on selected benchmarks.
KEY POINTS
- Mistral announced Voxtral 3, a family of speech-to-text and text-to-speech models plus a real-time voice agent SDK for multi-step conversations in 50+ languages; components include Voxtral Small Transcribe 3, Voxtral Mini TTS 3, and Pro Voice.
- The offering is available via Mistral’s EU-hosted API or for self-deployment, and Mistral says the Transcribe models achieve state-of-the-art diarization and multilingual transcription, claiming outperformance of Whisper, Scribe v2 and Gemini 3.5 Transcribe on selected benchmarks.
- Bundling SOTA transcription, expressive TTS, and a real-time agent SDK with EU-hosted or self-deploy options could materially lower the effort and compliance risk of deploying production voice agents.
DECISION BRIEF
CONFIDENCE
WHAT CHANGED
Mistral announced Voxtral 3, a new family of speech-to-text and text-to-speech models plus a real-time voice-agent SDK (voice agent that listens, reasons, calls tools, and speaks) with support for multi-step conversations.
WHY NOW
Voxtral 3 bundles state-of-the-art multilingual transcription, expressive TTS, and a streaming voice-agent SDK with EU-hosted API or self-deployment — potentially lowering integration and data-sovereignty friction for organizations building production voice agents.
WHO IS AFFECTED
Developers and vendors building voice agents and contact-center / virtual-assistant solutions; organizations requiring EU-hosted or self-deployed speech processing (data-sovereignty); users of multilingual transcription and expressive TTS in contact centers, service desks, virtual assistants, and hands-free copilots.
CONFIRMED
• Mistral introduced 'Voxtral 3', described as a family of speech-to-text and text-to-speech models plus a real-time voice agent SDK that combines listening, reasoning, tool calling, and speaking in a single streaming pipeline. (source: Mistral excerpt) • Named components include Voxtral Small Transcribe 3 (speech-to-text), Voxtral Mini TTS 3 (text-to-speech), and a Pro Voice offering. (source: Mistral excerpt) • Mistral states transcription covers 50+ languages; TTS supports over 35 languages and 50 dialects (Decoder excerpt reports TTS language/dialect counts). (sources: Mistral, Decoder) • Mistral says Voxtral Small Transcribe 3 provides diarization, context-biasing (keyphrase biasing), and end-of-turn detection. (source: Mistral excerpt) • The product is available via Mistral’s EU-hosted API or for self-deployment on customer infrastructure. (source: Mistral excerpt) • Mistral claims the Transcribe models are state-of-the-art in multilingual transcription/diarization and claims outperformance of Whisper, Scribe v2 and Gemini 3.5 Transcribe on selected benchmarks. (source: event summary / Mistral excerpt) • The Decoder article reports the SDK is based on LiveKit, that agents can call external systems during live conversations, and lists a transcription price starting at $0.0035 per minute. (source: Decoder excerpt)
UNCERTAIN
• Independent verification of the claimed benchmark outperformance (which benchmarks, evaluation methodology, and results) is not provided in the supplied excerpts — missing evidence. • Exact latency, throughput, and compute/hardware requirements for running Voxtral 3 (either via EU API or self-deploy) are not specified in the excerpts — missing evidence. • Details about 'Pro Voice' capabilities, licensing terms, commercial tiers beyond the single transcription price point, and data retention/processing specifics are not shown in the excerpts — missing evidence. • Real-world robustness across accents, noisy environments, and long multi-step dialogues is asserted but not independently demonstrated in the supplied excerpts — missing evidence. • Pricing information appears only in the Decoder excerpt (transcription $0.0035/min); the Mistral excerpt does not show full pricing — potential information gap between sources.
WHAT TO WATCH
Concrete observable signals to watch next: • Publication of detailed model cards or benchmark reports (methodology, benchmarks, and numbers). • Release of SDK docs, self-deploy artifacts or GitHub repo and LiveKit integration guides. • EU-hosted API availability pages, regional/contract terms, and data-retention/privacy documentation. • Independent third-party evaluations or comparative tests vs Whisper, Scribe v2, Gemini Transcribe. • Latency/throughput and compute-requirement benchmarks or example deploy configurations. • Additional pricing and commercial-tier disclosures beyond the reported $0.0035/min transcription price. • Early customer case studies, contact-center pilot announcements, or reference demos showing multi-step tool use in live calls.
WHY IT MATTERS
Bundling SOTA transcription, expressive TTS, and a real-time agent SDK with EU-hosted or self-deploy options could materially lower the effort and compliance risk of deploying production voice agents.
EVIDENCE MAP
4Editorial claims linked to specific sources, with support, contradiction and context shown separately.
Mistral released Voxtral 3, a family of speech-to-text and text-to-speech models plus a real-time voice-agent SDK for multi-step conversations in 50+ languages.
SUPPORTEDChecked 2026-10-20 · 2 supporting
Mistral's announcement describes Voxtral 3 as a family of STT/TTS models and a real-time voice-agent SDK supporting 50+ languages.
SUPPORTS The DecoderMEDIA · 2026-10-20The Decoder reports the same characterization of Voxtral 3 as a voice-agent stack covering 50+ languages.
Mistral says Voxtral Small Transcribe 3 is state-of-the-art for multilingual transcription, with diarization and context-biasing, and that it outperforms Whisper, Scribe v2 and Gemini 3.5 Transcribe on selected benchmarks.
SUPPORTEDChecked 2026-10-20 · 2 supporting
The Mistral page states that Voxtral Small Transcribe 3 is 'state-of-the-art' in multilingual transcription, includes diarization and context-biasing, and asserts benchmark superiority (details not provided in the excerpt).
SUPPORTS The DecoderMEDIA · 2026-10-20The Decoder reports Mistral's claim about state-of-the-art transcription and diarization; it repeats Mistral's benchmarking claim without showing the benchmarks in the excerpt.
Voxtral Mini TTS 3 provides expressive text-to-speech with emotional adaptation; The Decoder reports support for over 35 languages and 50 dialects.
SUPPORTEDChecked 2026-10-20 · 2 supporting
Mistral's page describes expressive speech and emotional adaptability for the Voxtral family, consistent with The Decoder's reporting (numbers appear in The Decoder excerpt).
SUPPORTS The DecoderMEDIA · 2026-10-20The Decoder states Voxtral Mini TTS 3 supports 35+ languages and 50 dialects with emotional adaptation.
Voxtral 3 is available via Mistral’s EU-hosted API or for self-deployment, positioning the product for sovereign deployment and customer control over where audio/data are processed.
SUPPORTEDChecked 2026-10-20 · 2 supporting
Mistral's announcement explicitly says Voxtral 3 is available via an EU-hosted API or for self-deployment to give organizations control over audio and data handling.
SUPPORTS The DecoderMEDIA · 2026-10-20The Decoder repeats that Voxtral can run on customers' infrastructure or via EU-hosted APIs for data control.
SOURCES & TIMELINE
2Voxtral 3 introduces a cutting-edge voice agent SDK and state-of-the-art speech models designed for real-time, multi-step conversations across 50+ languages. It combines accurate transcription, natural speech synthesis with emotional adaptability, and deep tool integration for seamless workflow automation. Available via Mistral’s EU-hosted API or self-deployment, it ensures data sovereignty while delivering low-laten…
Mistral AI has launched Voxtral 3, a new family of language models built for real-time voice agents. The system includes a speech-to-text model (Voxtral Small Transcribe 3), a text-to-speech model (Voxtral Mini TTS 3), and a voice agent SDK based on LiveKit. Together, they combine listening, thinking, tool use, and speaking into a single streaming pipeline. Transcription covers more than 50 languages. Text-to-speech …