Mistral has released Voxtral 3, a family of speech models available through the Mistral API, with transcription priced from $0.0035 per minute. The company announced the family on X at 00:00 Beijing time on October 21.
According to the price table in Mistral's announcement, Voxtral Mini Transcribe 3 costs $0.0035 per minute for batch transcription, Voxtral Small Transcribe 3 costs $0.0040 per minute, and Voxtral Mini Transcribe Realtime 3 costs $0.008 per minute. Voxtral Speech 3, the text-to-speech model, costs $13 per 1M characters for both batch and realtime use through the same API. Mistral says that price is lower than comparable models from OpenAI, Deepgram, ElevenLabs and Google.
Mistral says the transcription models cover 50 languages, including the 24 official EU languages, and support context biasing, where key terms can be supplied to improve recall on technical vocabulary. The announcement describes Voxtral Small Transcribe 3 as identifying who said what in a recording and detecting when a speaker has finished. Voxtral Mini Transcribe 3 has 4B parameters.
Voxtral Mini Transcribe Realtime 3 is built for voice agents and, according to Mistral, transcribes at under 500 milliseconds of latency in more than 35 languages, although a model card on the same page lists more than 40. Voxtral Speech 3 supports more than 35 languages and over 50 dialects, with controls for emotion, pacing and emphasis and voice cloning from samples of 30 minutes or more. Mistral's documentation names the speech models Voxtral TTS 3 and Voxtral TTS Realtime 3.
The release also includes a voice agent SDK built on LiveKit. It chains realtime transcription, a general model such as Mistral Small for reasoning and tool calls, and Voxtral Speech 3 into one streaming pipeline, and Mistral says each turn completes in under one second. Agents can handle calls over SIP and run in browsers and apps over WebRTC.
Mistral says Voxtral 3 can be used through its European, US and global API endpoints or deployed on a customer's own infrastructure. The announcement does not mention open weights. As of 00:50 Beijing time on October 21, Mistral's Hugging Face page listed only earlier Voxtral models.
Artificial Analysis said at 00:53 Beijing time on October 21 that Voxtral Small Transcribe 3 ranks third on its Speech to Text leaderboard with an AA-WER Index of 2.1%, ahead of ElevenLabs Scribe v2 at 2.2% and Grok Voice Transcribe 2.0 at 2.3%. Voxtral Mini Transcribe 3 ranks fifth at 2.2%. On streaming speech to text, Voxtral Mini Transcribe Realtime 3 scores 3.3% final-transcript AA-WER at the 960 ms setting, sixth of 46, and 3.6% at 480 ms. On the provider-voice text-to-speech arena, Voxtral 3 TTS scores an Elo of 1,218 across 1,599 appearances and ranks 12th of 98. On the controlled-voice arena, the same model scores an Elo of 1,068 across 1,614 appearances and ranks 12th of 43. On pronunciation robustness, Artificial Analysis lists 84.4%, 13th of 32 models.