Skip to content
CallMissed
Book a demo

Explore CallMissed

Back to all models
Speech to Textvoice-agentsstreamingturn-detectionmultilingual

Universal-3.6 Pro

by AssemblyAI · Released Undisclosed

AssemblyAI's newest streaming speech-to-text model, built for voice agents. Ends the caller's turn itself, auto-detects and code-switches across 32 languages including Hindi, Marathi, Urdu and Indian English. Voice sessions only. $0.4688 per audio hour.

Speech to Text

Universal-3.6 Pro

Powered by AssemblyAI · Streaming ASR with model-native turn detection (AssemblyAI Universal)

Context Window

N/A

Parameters

Undisclosed

Max Output

N/A

Category

Speech to Text

Overview

Universal-3.6 Pro is AssemblyAI's newest streaming speech recognizer, and on CallMissed it runs inside voice agents and voice sessions only. It decides when the caller has finished speaking, so a voice agent needs no separate turn detector on top of the transcript.

It auto-detects the spoken language and follows a caller who switches mid-sentence, so the language you set on the agent is a hint rather than a limit. The language list includes Hindi, Marathi and Urdu, plus Indian English (en-IN), which makes it a wider-coverage alternative to ink-2 for agents that take calls in more than one language.

It is streaming-only: POST /v1/audio/transcriptions returns an error for this model. For recorded audio use saaras:v4, ink-whisper or deepgram-nova-3.

Pricing

MetricPrice
Price /hour₹45.0000

1 credit = ₹1 ≈ US$0.0104 (US$1 = ₹96). Transparent per-call pricing — you pay only for what you call, with no seat fees or minimums.

Key Highlights

  • Model-native turn detection — no separate turn detector
  • 32 languages, including Hindi, Marathi, Urdu and Indian English
  • Auto-detects and follows mid-call language switches
  • Voice sessions only — not available for file transcription

Benchmarks

BenchmarkScore
Languages50
Turn detectionBuilt in
Price$0.4688
DeploymentStreaming only

Technical Details

  • Languages include English (with en-IN, en-US, en-GB, en-AU, en-CA, en-NZ), Hindi, Marathi, Urdu, Spanish, French, German, Italian, Portuguese, Arabic, Japanese, Chinese, Korean and more
  • Auto-detects the language and code-switches; the configured language is a hint
  • Streaming only — the file-transcription endpoint returns 400 for this model
  • Model-native turn detection (no separate turn detector required)
  • Selected per agent or voice session via stt_model

Strengths

  • Built-in turn detection removes a pipeline layer and its latency
  • Covers Hindi, Marathi, Urdu and Indian English in one model
  • Follows callers who switch language mid-call

Limitations

  • Streaming only — not available for file transcription (use saaras:v4 or ink-whisper for recorded audio)
  • Higher per-hour price than the Indic-specialist and Whisper-family models
  • Available inside voice agents and voice sessions only

Use Cases

Multilingual voice agentsIndian-language contact-centre automationCode-switching conversationsLive call transcription

API Example

curl https://api.callmissed.com/v1/voice/sessions \
  -H "Authorization: Bearer cm_YOUR_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "stt_model": "assemblyai-universal-3.6-pro",
    "language": "hi"
  }'

Endpoint: POST /v1/voice/sessions (streaming only) · Model ID: assemblyai-universal-3.6-pro

Try Universal-3.6 Pro now

Get 1000 free API credits on signup. No credit card required.

Start free Book a DemoRead docs