Sign in
Pending benchmarksTracking since Nov 18, 2026

Is AssemblyAI-STT-3.7-Realtime state of the art?Pending benchmarks.Announced Nov 17, 2026; no independent results yet.

AssemblyAI's streaming speech-to-text model in the AssemblyAI-STT-3.7 family, built for live conversations that end in a transcript: meetings, calls, captions, and notetakers. Official docs describe 32 languages in one stream with automatic detection, and speaker labels that are final the moment a turn ends, with no revision pass. It replaces Universal-Streaming and Universal-Streaming Multilingual. Live on AssemblyAI's Realtime Speech-to-Text API as speech_model="assemblyai-stt-3-7-realtime" at $0.30 per hour of session time. Official docs say it is not tuned for agent turn-taking. No weights. Official models.md dates the family launch to November 17, 2026; the changelog and lab blogs are dated November 18.

Announced facts

The details so far

Lab
AssemblyAI
Announced
Nov 17, 2026
API
Live · assemblyai-stt-3-7-realtime · Realtime Speech-to-Text API: WebSocket wss://streaming.assemblyai.com/v3/ws with speech_model="assemblyai-stt-3-7-realtime" and an API key. Official introducing post: the endpoint and message protocol are unchanged from the previous streaming models; speaker labels no longer send a SpeakerRevision message
Open weights
Closed · Not released
Task
Streaming speech-to-text for live conversations: meetings, calls, captions, notetakers, and agent assist. Official docs: not tuned for agent turn-taking
Languages
32 languages in one stream, with automatic language detection (official introducing post, models reference, and changelog)
Pricing
$0.30 per hour of session time (official pricing reference and introducing post). Official Voice Focus docs: Voice Focus is not available on this model
Speaker labels
Official introducing post: from the start of a new session, speaker labels are final the moment a turn ends and are nearly as accurate as labels from pre-recorded audio; there is no revision pass. Partials arrive about once a second
Availability
Generally available on AssemblyAI's self-serve Realtime Speech-to-Text API. Official changelog dated 2026-11-18. It replaces Universal-Streaming and Universal-Streaming Multilingual, which official docs mark deprecated as of November 17, 2026 (tentative retirement February 1, 2027)
Related
Separate from AssemblyAI-STT-3.7 (pre-recorded) and from AssemblyAI-STT-3.7-Realtime-Agent (voice agents), from AssemblyAI-TTS-1, and from Universal-3.5 Pro

Tags marked unsure are not stated on the lab's own pages.

Next steps

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

  • Publicly announced with a source
  • Usable by the public (API or weights)
  • Listed on an independent board we track
  • Enough results for a rank in a category
  • Holds across two snapshots (crowns only)
Self-reported

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: measured on the same recordings, AssemblyAI reports a lower live diarization error rate than the other realtime providers it measured on CallHome (40 unscripted telephone calls) and on NOTSOFAR (129 real meetings recorded on a single distant microphone, up to seven speakers), including ElevenLabs Scribe v3 Live and Meta Muse Voice Transcribe (self-reported, official introducing post).

Receipts

Sources

LABAssemblyAI models reference (llms/models.md)assemblyai.com/llms/models.md · Nov 17, 2026LABSpeech model versions and deprecationsassemblyai.com/docs/getting-started/model-versions-and-deprecations · Nov 17, 2026LABAssemblyAI changelog: AssemblyAI-STT-3.7assemblyai.com/changelog/ · Nov 18, 2026LABIntroducing AssemblyAI-STT-3.7 (AssemblyAI blog)assemblyai.com/blog/introducing-assemblyai-stt-3-7 · Nov 18, 2026LABStreaming model selectionassemblyai.com/docs/streaming/select-the-speech-modelLABAssemblyAI pricing referenceassemblyai.com/pricing.mdLABAssemblyAI on X: AssemblyAI-STT-3.7x.com/AssemblyAI/status/2123080177508929849 · Nov 18, 2026

Tracking continues until an independent board lists it.