Daily AI CatchupMistral and ElevenLabs Launch Rival Voice Agent Models
Mistral released Voxtral 3, a suite covering transcription, text-to-speech, and a full voice agent SDK across 50+ languages. The same day, ElevenLabs launched Eleven Realtime, speech-to-speech models the company says can reason while speaking rather than just converting text. Both aim at the same target: businesses building phone and web voice agents, a category that's quickly getting crowded.
Why it matters
💻 Developer · Two production-ready voice SDKs with multilingual support means fewer reasons to build your own pipeline from ASR plus LLM plus TTS components.
📦 Product · Voice agents for phone support or booking are now a buy-vs-build decision with real vendor options, not a research project.
🎨 Design · Speech-to-speech models that handle emotion and reasoning mid-conversation change what a 'natural' voice interface can sound like.
📈 Business · Competing voice agent platforms from Mistral and ElevenLabs should push prices down for call-center and customer-service automation.
🤔 Just Curious · The robotic phone-bot voice is getting replaced by something closer to a real conversation, for better or worse.
Sources: Voxtral 3: Voice Agent SDK & Speech Models in 50+ Languages, Eleven Realtime