AssemblyAI Launches STT-3.7 Models With 86ms Agent Latency and Rewritten Inference Engine

AssemblyAI · x · 2026-11-19

AssemblyAI shipped three STT-3.7 transcription models and its first TTS model after rewriting its inference engine. The Realtime-Agent model hits 86ms final-transcript latency (2x faster), ranks #1 in accuracy for noisy telephony voice-agent calls, and all models cover 32 languages; a 5-minute file transcribes in 9 seconds.

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

Original post →

More from Multimodal

Multimodal channel →