Universal-3.6 Pro Realtime
Audio modelCard summarizedOur flagship streaming model.
Source: Universal-3.6 Pro Realtime launch post and AssemblyAI models reference (no model card published)
Where to call it
- AssemblyAI Streaming API
universal-3-6-pro
Good for
- Real-time voice agents needing accuracy, broad language support and conversational context
- Short confirmations (yes/no) and entities such as numbers, codes, emails and addresses
- Noisy rooms, accented speech and phone lines
Overview
AssemblyAI · 01Pro, front
Universal-3.6
Our flagship streaming model.
Typical price
$0.45 / hour
Streaming, base model
- Max length
- 3 h per streaming session
- Languages
- 32 languages with native code-switching
- Modes
- speech-to-text, streaming
- Released
- 29 Sep 2026
Universal-3.6 Pro Realtime is AssemblyAI's streaming speech-to-text model for voice agents and a direct upgrade to Universal-3.5 Pro Realtime, at the same price and latency. AssemblyAI says it was trained on tens of thousands of hours of real voice-agent and telephony conversations, cuts wrong short confirmations by about 45%, holds quiet-call accuracy in noise, accents and phone audio, and serves 32 languages from one model. launch post; models.md
universal-3-6-pro- Typical price
- $0.45 / hour
- Max length
- 3 h per streaming session (auto-closes)
- Languages
- 32 languages with native code-switching (14 new vs 3.5 Pro)
- Weights
- Closed
- Released
- 29 Sep 2026
- speech-to-text
- streaming
- end-of-turn detection
- native code-switching
- automatic language detection
- keyterms prompting
- prompting (beta)
- real-time speaker diarization
- Medical Mode
- entity recognition