Mistral Voxtral 3 Adds Batch, Realtime Models

Voxtral 3 turns Mistral’s speech stack into a more complete developer platform, with stronger transcription claims and agent-ready streaming primitives—but at higher prices than Voxtral 2.

• –Mini Transcribe 3 ranks third on Mistral’s cited FLEURS multilingual transcription and English diarization benchmarks, surpassing Gemini 3.5 Transcribe.
• –Realtime 3 combines WebSocket streaming, end-of-turn detection, speaker diarization, and context biasing in one session.
• –The new aliases provide a straightforward migration path from `voxtral-mini-2602` and `voxtral-mini-transcribe-realtime-2602`. [Migration guide](https://docs.mistral.ai/studio/audio/migrate-from-voxtral-2)
• –Pricing rises from $0.003 to $0.0035 per minute for batch transcription and from $0.006 to $0.008 for realtime, making the upgrade worthwhile mainly for applications needing the new capabilities.