**[Mistral Releases Voxtral 3 Voice Agent SDK Supporting 50+ Languages]** According to a report by Jinse Finance on October 21, Mistral AI officially announced the release of Voxtral 3—a voice agent SDK and voice model family designed for real-time multi-step voice conversations, supporting over 50 languages. It integrates accurate transcription, emotionally expressive natural speech synthesis, and deep tool invocation into a single real-time streaming pipeline, applicable to call centers, virtual assistants, and hands-free copilots. Voxtral 3 offers two deployment options: EU-hosted API or self-hosted, emphasizing data sovereignty and low latency. Its speech transcription model (Voxtral Small Transcribe 3) supports speaker separation and contextual biasing, claiming to outperform Whisper, Scribe v2, and Gemini 3.5 Transcribe. The speech synthesis model (Voxtral Mini TTS 3) covers 35+ languages and 50+ dialects, priced at $13 per million characters, reportedly cheaper than similar offerings from OpenAI and ElevenLabs.