--- title: Mistral launches Voxtral 3 speech models and a voice... description: Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models alongside an SDK for building voice agents. The SDK handles listening,... canonical: https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Mistral launches Voxtral 3 speech models and a voice agent SDK | daily.dev og:description: Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models alongside an SDK for building voice agents. The SDK handles listening,... og:url: https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp og:image: https://api.daily.dev/og/posts/OGvbet0zp.png og:image:alt: Mistral launches Voxtral 3 speech models and a voice agent SDK og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Mistral launches Voxtral 3 speech models and a voice agent SDK **[Collections](https://daily.dev/sources/collections)** · 1 min read · 2 upvotes · 0 comments ## Summary Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models alongside an SDK for building voice agents. The SDK handles listening, reasoning, tool calling and speaking in one real-time pipeline and powers Mistral's new voice agents API. Transcription models cover more than 50 languages with diarization and context biasing; the Mini model costs $0.0035 per minute, and a realtime transcription model hits sub-500ms latency across 35+ languages. Text-to-speech covers 35+ languages and 50+ dialects at $13 per 1M characters, undercutting OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. The SDK runs on LiveKit, completes each turn in under a second, and is available via Mistral's EU-hosted API or can be deployed yourself. ## Content Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models plus an SDK for building real-time voice agents. The SDK handles listening, reasoning, tool calling and speaking in one pipeline. ## Transcription Voxtral Small Transcribe 3 supports 50+ languages, including 24 EU languages and low-resource Asian and Indic languages. It handles diarization and context biasing. Mistral says it outperforms Whisper, Scribe v2 and Gemini 3.5 Transcribe. On the Artificial Analysis Speech-to-Text leaderboard it landed in 3rd place. The Mini variant costs $0.0035 per minute. A separate realtime transcription model reaches sub-500ms latency across 35+ languages. ## Text-to-speech Voxtral Mini TTS 3.0 covers 35+ languages and 50+ dialects, and adapts emotion and dialect to the conversation. It ranked in the top 12 on the Artificial Analysis leaderboard. Pricing is $13 per 1M characters, which undercuts OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. A Pro Voice option offers more accurate voice adaptation. ## Voice agent SDK The SDK is built on LiveKit. Each conversational turn completes in under one second. ## Availability Voxtral 3 models are available through Mistral's EU-hosted API or can be self-deployed. ## Questions this post answers ### How much does Mistral's Voxtral 3 transcription cost per minute? The Mini transcription model costs $0.0035 per minute. The Voxtral 3 family supports more than 50 languages with diarization and context biasing, and a separate realtime transcription model reaches sub-500ms latency across 35+ languages. _daily.dev helps developers pricing out voice agent stacks keep tabs on releases like Voxtral 3._ ### How much does Mistral's Voxtral 3 text-to-speech cost compared to ElevenLabs or OpenAI TTS-1? Voxtral 3 text-to-speech is priced at $13 per 1M characters, which undercuts OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. It covers 35+ languages and 50+ dialects. _Teams comparing TTS vendors on cost can track pricing shifts like this one on daily.dev._ ### What does Mistral's voice agent SDK use for real-time conversation handling? The SDK runs on LiveKit and handles listening, reasoning, tool calling and speaking in a single real-time pipeline, completing each conversational turn in under one second. It can be used through Mistral's EU-hosted API or deployed self-hosted. _Developers building voice agents can follow SDK releases like this one on daily.dev._ ## Community take How the wider developer community reacted, aggregated from 2 discussions and 245 comments across hackernews, x (as of 2026-10-20). **TL;DR:** People are impressed by Voxtral 3's pricing, multilingual coverage, and low-latency turn-taking, but a loud contingent is frustrated that it isn't open-weight like earlier Voxtral releases, and many are asking pointed technical questions about barge-in latency and noisy-audio accuracy that the announcement doesn't address. Voxtral 3: The main point of contention is that Voxtral 3 is closed-weight (unlike the prior Voxtral generation), which disappoints people who wanted local/self-hostable speech models, sparking a long thread of alternative open-source STT/TTS recommendations; opinions on the actual voice quality and real-time pipeline are mixed to negative. **Sentiment:** 31% positive · 31% mixed · 38% skeptical **The case for** - Pricing undercuts competitors noticeably, e.g. transcribing an hour of audio for about 24 cents. - Diarization and context biasing are seen as genuinely useful additions for custom ASR pipelines. - The unified listen-reason-speak pipeline on LiveKit is seen as a practical, builder-friendly architecture. - Some found the earlier Voxtral series' transcription and diarization quality already quite good and cheap. - Sub-second turn-taking and end-of-speech detection are called out as the quiet but important headline feature. **The pushback** - Repeated frustration that the models aren't open-weight, breaking from prior Voxtral releases. - It's closed/not open-weight, unlike the previous Voxtral generation, which disappoints people who want local, privacy-preserving deployment. - Skepticism that transcription accuracy claims will hold up on messy real-world audio (overlapping speakers, background noise, accents) versus clean benchmark conditions. - Several listeners feel the new TTS voices sound more artificial and 'AI slop' compared to the more natural-sounding version 1 demos. - Doubts about whether barge-in/interruption handling and end-of-utterance detection actually work well in practice. **By community** - hackernews (mixed): A vocal contingent is frustrated the model is closed and shares a wide range of open local alternatives, while others note real improvements in diarization and latency, with some disappointed in the new voice quality specifically. - x (mixed): Enthusiasm about pricing, latency, and language coverage is offset by persistent complaints about the lack of open weights and unanswered questions about real-world turn-taking performance. **Hottest debate:** Whether Mistral's decision to keep Voxtral 3 closed (unlike earlier open-weight Voxtral models) is a betrayal of its open-source reputation or a reasonable enterprise move. **Open questions** - How does the system actually handle mid-sentence interruptions and barge-in in practice, not just in demos? - Will Mistral eventually release smaller, locally deployable versions of these models? - Does transcription accuracy hold up with overlapping speakers, background noise, and mixed-language sentences? - How does Voxtral 3's diarization actually compare against Nvidia's Nemotron 3 Diarization model? - Will Mistral ever release open weights for Voxtral 3 or the full voice agent stack? **Highlights** > @MistralAI But why no open weight models? > — [coder543 on x · 1 points](https://x.com/coder543/status/2112587136889504158) > @MistralAI Curious what's behind the state-of-the-art claim: which benchmarks, and does it hold up on the hard stuff - overlapping speakers, noisy rooms - not just clean studio audio? Transcription leaderboards tend to reward the easy cases. > — [BMTestsAI on x · 1 points](https://x.com/BMTestsAI/status/2112610365528031635) > @MistralAI Mistral just priced transcribing an hour of audio at 24 cents. ElevenLabs has a pricing problem now. > — [nagasankey on x · 1 points](https://x.com/nagasankey/status/2112578889067175982) > @MistralAI I don't get the point of releasing this. It's super expensive compared to any OS model like Parakeet running on or Replicate. By huge margins > — [kingdrale on x](https://x.com/kingdrale/status/2112588801369633277) > @MistralAI Transcription WER is table stakes now — what decides a voice-agent stack is turn-taking: barge-in latency, backchannels, end-of-utterance detection. Where does Voxtral 3 land on time-to-first-audio when a caller interrupts mid-sentence? > — [ReignSwords on x · 1 comments](https://x.com/ReignSwords/status/2112606581657645516) **Source threads** - [x](https://x.com/MistralAI/status/2112574615318876178) · 0 points · 175 comments - [hackernews](https://news.ycombinator.com/item?id=50168396) · 245 points · 70 comments ## Similar posts on daily.dev - [Voxtral-Mini-4B-Realtime for real-time speech transcription is now available in Amazon SageMaker JumpStart](https://daily.dev/posts/voxtral-mini-4b-realtime-for-real-time-speech-transcription-is-now-available-in-amazon-sagemaker-jum-y4vpyj230) · AWS · 0 upvotes · 0 comments - [Run Voxtral Mini 4B Realtime on vLLM with Red Hat AI on Day 1](https://daily.dev/posts/run-voxtral-mini-4b-realtime-on-vllm-with-red-hat-ai-on-day-1-wyohmx0fj) · Red Hat Developer · 0 upvotes · 0 comments --- Tags: [#speech-recognition](https://daily.dev/tags/speech-recognition), [#text-to-speech](https://daily.dev/tags/text-to-speech), [#voice-ai](https://daily.dev/tags/voice-ai), [#mistral-ai](https://daily.dev/tags/mistral-ai) [View this post on daily.dev](https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Mistral launches Voxtral 3 speech models and a voice agent SDK","url":"https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp"},"datePublished":"2026-10-20T17:11:40.292Z","dateModified":"2026-10-20T20:49:11.768Z","description":"Mistral has released Voxtral 3, a family of speech-to-text and text-to-speech models alongside an SDK for building voice agents. The SDK handles listening,...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d017db232b20c2e3a158103d9a8fae5?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/0d017db232b20c2e3a158103d9a8fae5?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"Collections","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"Collections","logo":"https://media.daily.dev/image/upload/s--fk_6ycEi--/f_auto,q_auto/v1780996001/logos/collections?_a=BAMAMiWQ0","url":"https://daily.dev/sources/collections"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":2},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"speech-recognition,text-to-speech,voice-ai,mistral-ai","timeRequired":"PT1M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Collections","item":"https://daily.dev/sources/collections"},{"@type":"ListItem","position":3,"name":"Mistral launches Voxtral 3 speech models and a voice agent SDK"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/mistral-launches-voxtral-3-speech-models-and-a-voice-agent-sdk-ogvbet0zp#faq","mainEntity":[{"@type":"Question","name":"How much does Mistral's Voxtral 3 transcription cost per minute?","acceptedAnswer":{"@type":"Answer","text":"The Mini transcription model costs $0.0035 per minute. The Voxtral 3 family supports more than 50 languages with diarization and context biasing, and a separate realtime transcription model reaches sub-500ms latency across 35+ languages. daily.dev helps developers pricing out voice agent stacks keep tabs on releases like Voxtral 3."}},{"@type":"Question","name":"How much does Mistral's Voxtral 3 text-to-speech cost compared to ElevenLabs or OpenAI TTS-1?","acceptedAnswer":{"@type":"Answer","text":"Voxtral 3 text-to-speech is priced at $13 per 1M characters, which undercuts OpenAI TTS-1, Deepgram, ElevenLabs Flash and Gemini Flash. It covers 35+ languages and 50+ dialects. Teams comparing TTS vendors on cost can track pricing shifts like this one on daily.dev."}},{"@type":"Question","name":"What does Mistral's voice agent SDK use for real-time conversation handling?","acceptedAnswer":{"@type":"Answer","text":"The SDK runs on LiveKit and handles listening, reasoning, tool calling and speaking in a single real-time pipeline, completing each conversational turn in under one second. It can be used through Mistral's EU-hosted API or deployed self-hosted. Developers building voice agents can follow SDK releases like this one on daily.dev."}}]} ```