Tasks and inputs
AssemblyAI-STT-3.7 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 is AssemblyAI's November 2026 pre-recorded transcription default: 32 languages, new diarization, fewer language flips, $0.21 per hour.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Indicative: $0.21 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
Speaker labels are where batch transcription usually disappoints. The headline of AssemblyAI-STT-3.7, released on 17 November 2026, is a new diarization model, and the numbers move a lot: concatenated-permutation WER fell from 29.3% to 24.0%, diarization error rate from 14.7% to 10.4%, and average speaker-count error from 0.57 to 0.20. On short 60-second clips, cpWER dropped from 34.0% to 19.1%.
AssemblyAI-STT-3.7 (assemblyai-stt-3-7) replaces Universal-3.5 Pro as the recommended model for recorded audio, at the same $0.21 per hour. Batch recognisers like this one sit alongside open options such as Whisper Large v3 Turbo in the speech to text category.
The model handles 32 languages, up from 18. Additions include Urdu, Russian, Korean, Catalan, Galician, Romanian, Estonian, Persian, Cantonese, Afrikaans, Marathi, Zulu, Xhosa and Norwegian Nynorsk. Vendor-reported WER on some of the new ones: Galician 4.1%, Russian 4.6%, Catalan 5.1%. Across the 17 non-English languages carried over from the previous model, mean error went from 12.8% to 10.1%.
Code-switched audio, a speaker sliding between two languages, now triggers 80 to 90% fewer spurious language flips.
| Setting | What to know |
|---|---|
| Parameter | speech_models: ["assemblyai-stt-3-7"] |
disfluencies |
Defaults to true (keeps "um", "uh") |
format_text |
Defaults to true |
| Fallback | None unless you list universal-2 as a second model |
| Response | speech_model_used tells you which model ran |
Defaults differ from Universal-3.5 Pro, so set both transcript-style flags explicitly when you switch or your downstream text will change shape. For languages outside the 32, add Universal-2 to the list; it covers 99.
Other features carry over: keyterms up to 1,000 terms, Medical Mode for English, Spanish, German and French, PII redaction and standard diarization. Prompting and keyterms cannot be combined on the same async request. Experimental diarization is not offered on this model.
Universal-3 Pro and Universal-3.5 Pro (pre-recorded) are now deprecated, with retirement dates to be announced. Once retired, requests naming them route to STT-3.7 automatically. That is convenient, and also a reason to test now: your output will change on a date you do not control.
Re-run a month of already-transcribed calls with known speaker counts. Score WER, cpWER and the share of files where the speaker count is wrong. Then send your hardest code-switched recordings separately. This release went live on 17 November 2026, so independent benchmarks have not caught up; every figure above is AssemblyAI's own. Pricing is indicative. For streaming needs, see the separate Realtime and Realtime-Agent variants, and compare other recognisers under automatic speech recognition.
Compare the model on the work you need to complete.
AssemblyAI-STT-3.7 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-Realtime streams transcripts in 32 languages for captions, notetakers and agent assist, with final speaker labels per turn at $0.30/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.