Tasks and inputs
Universal-3.6 Pro is listed for Audio, Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
AssemblyAI Universal-3.6 Pro is a realtime streaming speech-to-text model covering 32 languages with code switching, at $0.45 per audio hour.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: realtime streaming $0.45 per audio hour (volume discounts); a sync/pre-recorded variant is listed at ~$0.000083 per audio second (~$0.30/hour) with a higher prompted rate. Confirm on your account.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Voice agents live or die on two things: hearing names right and knowing when the user has stopped talking. AssemblyAI Universal-3.6 Pro, launched on 29 September 2026, targeted both. It is a realtime streaming upgrade to Universal-3.5 Pro, and moving to it only requires changing the model name.
Before anything else, though, note its status. As of 17 November 2026, AssemblyAI lists Universal-3.6 Pro as previous generation. New builds are pointed to the AssemblyAI-STT-3.7 realtime models at the same $0.45 per hour. Existing integrations keep working; new ones should evaluate 3.7 first.
Language coverage was the big change. One model handles 32 languages with automatic detection and switches language mid-stream when speakers do. Fourteen were new in this release, including Korean, Russian, Urdu, Zulu and Cantonese.
On AssemblyAI's own test of 12,460 English voice-agent scenarios:
The entity figure is the one to watch. Names, account numbers and addresses are what break agent workflows, and one in seven entities still comes out wrong on the vendor's benchmark.
Keyterm prompting accepts up to 100 terms and can be updated mid-stream, so you can inject a caller's name once you know it. Medical Mode covers four languages. Real-time speaker diarization is an add-on.
The model name also appears with a pre-recorded, synchronous variant. That variant is listed at $0.000083 per audio second, about $0.30 per hour, with a higher rate for prompted requests. The realtime streaming model is $0.45 per hour, with volume discounts. These are different endpoints with different prices; do not mix up the figures. A dated listing slug suggests the sync variant arrived around 14 October 2026. All prices are indicative, so confirm on your account.
| Item | Detail |
|---|---|
| Model name | universal-3-6-pro |
| Launched | 29 September 2026 |
| Mode | Realtime streaming (sync variant also listed) |
| Languages | 32, auto-detected, code switching |
| Streaming price | $0.45 per audio hour |
| Status | Previous generation |
AssemblyAI's Dictation API also runs on this model.
Only if you are already on it and stable. Otherwise, record fifty real calls, including accented speakers and mid-sentence language switches, and run them through 3.6 Pro and STT-3.7 side by side. Score entity accuracy by hand on the twenty calls that matter most. For alternatives, see speech recognition models or compare with Gemini 3.5 Transcribe Live.
Compare the model on the work you need to complete.
Universal-3.6 Pro is listed for Audio, Automatic speech recognition, Speech → Text. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
salesforce
SOV-Transcribe-1 is Salesforce's open-weight speech-augmented model for non-streaming transcription, with keyword biasing and style instructions; Apache-2.0.
ElevenLabs
Scribe v3 Live is ElevenLabs' streaming transcription model with turn prediction, mid-sentence language switching and a 10,000-term keyterm library.
NVIDIA
Nemotron 3.5 Transcribe 30B A4B is NVIDIA's open offline ASR model: 40 languages, hour-long files, word timestamps, diarisation and translation.
IBM
Granite Speech 5.0 2B is IBM's open speech-language model for multilingual ASR and translation with keyword biasing, runnable from Ollama to MLX.
Mistral AI
Voxtral Mini Transcribe 3 is Mistral's 4B batch transcription model with speaker labels, segment timestamps and custom vocabulary, replacing Mini 2.
Mistral AI
Voxtral Mini Transcribe Realtime 3 is Mistral's streaming speech-to-text model for voice agents: under 500 ms, tunable delay, built-in turn detection.