Mistral's Voxtral 3 prices speech-to-text at $0.0035 a minute

The lineup

Mistral's announcement lists Small Transcribe 3 (top accuracy), Mini Transcribe 3 (4B parameters), Mini Transcribe Realtime 3 (sub-500ms latency) and Mini TTS 3, plus a voice-agent SDK built on LiveKit. Transcription covers 50+ languages including all 24 official EU languages; realtime 35+; speech output 35+ languages.

Vendor-reported results: state of the art on FLEURS multilingual WER ahead of Whisper large-v3, Scribe v2 and Gemini 3.5 Transcribe, and about 18.4% cpWER on diarization. The voice agent API costs $0.04/min with Mistral Small 4. The page states no license.

Accuracy is not the gap, and the weights are not coming

A developer who has built voice agents for the past year says word error rate is already commodity, so the benchmarks Mistral leads with miss the hard part:

"The real gaps are in latency, handling interruptions, interrupting the speaker if needed… and adapting to the emotion/tone of the context."

On the license, a commenter who checked Hugging Face found the answer is no:

"The post offers the API, and self-deployment goes through sales. Nothing for Voxtral 3 is on Hugging Face. The last open one is Voxtral Mini 4B Realtime from February, Apache 2.0"

That older model is what a developer who launched commercial offline speech-to-text software on it builds on, running on a 4GB-VRAM laptop iGPU. For anyone needing local or private transcription, the $0.0035/min price is for a hosted service only, and the open Mini 4B Realtime remains the self-hosted option.