Tasks and inputs
AssemblyAI-STT-3.7-Realtime is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. It accepts audio and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 18, 2026
AssemblyAI-STT-3.7-Realtime streams transcripts in 32 languages for captions, notetakers and agent assist, with final speaker labels per turn at $0.30/hour.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.30 per audio hour. Confirm on your AssemblyAI 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
Not every live transcript has someone waiting on a reply. Meeting notetakers, live captions, call coaching and agent-assist screens all need text quickly, but none of them must decide the instant a speaker finishes. AssemblyAI built a separate streaming model for that job and priced it below its voice-agent variant.
AssemblyAI-STT-3.7-Realtime (assemblyai-stt-3-7-realtime) launched on 17 November 2026 at $0.30 per hour.
It replaces both Universal-Streaming English and Universal-Streaming Multilingual. All 32 supported languages run in a single stream with automatic detection, so a deployment that used a language router in front of two separate integrations can drop the router.
Set speaker_labels: true and each turn's label is final when the turn ends. There is no later revision pass. AssemblyAI tested this on five meeting and call datasets, 466 recordings and 83 hours, against Universal-3.6 Pro Realtime:
| Metric | STT-3.7-Realtime | Universal-3.6 Pro Realtime |
|---|---|---|
| DER | 11.67% (live) | 21.17% live / 14.96% revised |
| cpWER | 24.25% | 35.31% live / 25.84% revised |
Live labels from the new model beat even the older model's revised ones.
Versus the same predecessor: long-form English 8.56% WER (from 9.03%), short-form English 3.22% (from 3.44%), a 30-language average of 11.72% (from 12.25%) and 7.35% across five code-switching pairs (from 7.85%). Modest gains; the diarization change is the bigger story.
wss://streaming.assemblyai.com/v3/ws, same message protocol.prompt.That ignored-parameter line is the important one. Turns close when enough content has arrived, not on fast end-of-turn decisions, which is why AssemblyAI says not to put this model behind a voice agent. Use the Realtime-Agent variant for that.
Shadow your current captioning or notes stream with this model for a fortnight. Compare speaker-attribution errors in the summaries people actually read, plus WER on a hand-checked sample. All numbers here are vendor-reported from launch day, and prices are indicative. Batch work belongs on the pre-recorded AssemblyAI-STT-3.7 model. For other live recognisers, compare Gemini 3.5 Transcribe Live and the wider automatic speech recognition category.
Compare the model on the work you need to complete.
AssemblyAI-STT-3.7-Realtime is listed for Automatic speech recognition, Speech → Text, Speaker Diarization. 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.
assemblyai
AssemblyAI-STT-3.7 is AssemblyAI's November 2026 pre-recorded transcription default: 32 languages, new diarization, fewer language flips, $0.21 per hour.
assemblyai
AssemblyAI-STT-3.7-Realtime-Agent is a streaming recogniser for voice agents and IVR: 32 languages, 86 ms to final transcript, 4.46% vendor WER, $0.45/hour.
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.
NVIDIA
Nemotron-Labs-Unified-ASR-0.6b is NVIDIA's open 600M speech recognizer for 25 European languages, running offline or streaming at down to 160 ms.