Languages supported by Universal-3.6 Pro Realtime for voice agents

Universal-3.6 Pro Realtime is a multilingual streaming speech-to-text model for voice agents that supports 32 languages, including 13 newly added languages such as Cantonese, Korean, Russian, Urdu, Xhosa, and Zulu. It natively handles code-switching, including changes within a sentence, preserving words in their spoken languages rather than translating them, and is reported to achieve a 7.20% word error rate on code-switched speech compared with 17.13% for Nova-3 Multi. Developers can optionally enable per-turn language detection, which returns a language code and confidence score without affecting transcription, or steer recognition toward one or more expected languages through configuration that can be updated during a session. The model is the default on the streaming endpoint, integrates with platforms such as LiveKit and Pipecat, and costs $0.45 per hour billed by the second, while an agent’s reply language remains dependent on its LLM and text-to-speech system; AssemblyAI plans to launch its Text-to-Speech API on October 28, 2026.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.