Mistral just shipped Voxtral 3 — a complete voice AI stack that's actually usable.
What's in it:
• Transcription in 50+ languages with speaker labels + custom keywords
• Real-time transcription under 500ms latency
• Text-to-speech in 35+ languages + voice cloning
• Voice agent SDK that listens, reasons, calls tools, and responds
Mistral claims ~4% word error rate vs OpenAI Whisper large-v3's ~16%. Vendor numbers, so take with a grain of salt until independent tests drop.
Pricing starts at $0.0035/min for transcription.
Voice agents just got way cheaper to ship. If you're building in AI agents or consumer voice apps, this is worth testing immediately.
What's in it:
• Transcription in 50+ languages with speaker labels + custom keywords
• Real-time transcription under 500ms latency
• Text-to-speech in 35+ languages + voice cloning
• Voice agent SDK that listens, reasons, calls tools, and responds
Mistral claims ~4% word error rate vs OpenAI Whisper large-v3's ~16%. Vendor numbers, so take with a grain of salt until independent tests drop.
Pricing starts at $0.0035/min for transcription.
Voice agents just got way cheaper to ship. If you're building in AI agents or consumer voice apps, this is worth testing immediately.