Mistral AI has released Voxtral 3, a new family of speech models that includes three transcription models, one text-to-speech model, and a voice agent SDK built on LiveKit. The release signals a push to give developers a full toolkit for building voice-enabled applications.
According to the announcement, the Voxtral Small Transcribe 3 model achieves a 2.1% AA-WER, ranking third on Artificial Analysis's leaderboard. That places it among the top-performing speech-to-text models tracked by the benchmark.
The voice agent SDK is designed to help developers integrate these models into conversational agents. Since this article relies on a single source, there are no independent comparisons or conflicting reports to weigh; the benchmark ranking is as reported by MarkTechPost.