Updated

Universal-3.6 Pro Realtime

AssemblyAI released Universal-3.6 Pro Realtime, an upgrade to Universal-3.5 Pro for streaming speech recognition. It supports 32 languages with automatic detection and is designed to handle short responses, multiple speakers, accents, and noisy telephony audio at the previous version’s latency.

By No Code MBA · Facts checked

Provider
AssemblyAI
Announcement date
Status at last check
Generally available
Model family
Universal
Preceding model
Universal-3.5 Pro Realtime
Capabilities
Audio, Text

What changed

Universal-3.6 Pro Realtime is a streaming speech-to-text upgrade trained on voice-agent and telephony conversations. It covers 32 languages (14 new, including Korean, Russian and Cantonese) with automatic detection at the same 537 ms median endpoint latency. AssemblyAI reports wrong confirmations on short answers fell from 2.65% to 1.45%, code-switching WER fell from 8.65% to 7.20%, and endpointing precision rose from 74.1% to 77.9%. On its English voice-agent benchmark it reports 5.19% WER.

Access and availability

Use speech_model=universal-3-6-pro on the streaming endpoint wss://streaming.assemblyai.com/v3/ws; no other request changes are needed. Universal-3.5 Pro Realtime remains available for comparison.

Official access documentation

Pricing

$0.45 per hour of audio, with volume discounts

Official pricing

What you can build

Test it on a representative set of noisy calls with short confirmations, names, and code-switching, and measure transcription errors before connecting its output to automated actions.

Limitations

Most headline numbers come from AssemblyAI's own benchmark with competitors at default settings, and code-switching was tested only on five English to European language pairs. AssemblyAI notes Meta's model was more accurate but slower in Pipecat's independent run.

Sources and release history