AssemblyAI launches STT-3.7 as three transcription models for calls, agents and recordings

AssemblyAI · x · 2026-11-18

AssemblyAI released its next-gen transcription models, AssemblyAI-STT-3.7, shipped as three specialized models: Realtime for live conversations with first-time-right speaker labels and lowest live diarization error, Realtime-Agent for voice agents that need final transcripts before the caller notices a pause, and a third model for recordings where accuracy and speaker attribution are scored together. The thesis: meetings, agents and recordings fail in different ways, so each model optimizes for one core need.

Related event: AssemblyAI Launches STT-3.7 Transcription Models for Real-World Audio(3 posts)→

Original post →

More from Models

Models channel →