Updated

AssemblyAI model releases

AssemblyAI model announcements from September 2026 to November 2026, newest first. Compare what changed and follow the official sources.

Latest releases

Dates refer to the linked announcement or release milestone. Previews are labeled in the entries; historical availability and pricing may have changed.

Release details

  1. AssemblyAI

    AssemblyAI-STT-3.7

    AssemblyAI released AssemblyAI-STT-3.7 for recorded audio transcription. The model supports 32 languages with automatic detection and an optional 99-language fallback, plus context biasing and natural-language formatting instructions.

    For buildersTest transcription on representative recordings with domain vocabulary, then review names and numbers before using the output in downstream workflows.

  2. AssemblyAI

    AssemblyAI-STT-3.7-Realtime

    AssemblyAI released AssemblyAI-STT-3.7-Realtime for live conversations, including meetings, calls, and captions. It supports 32 languages in one stream with automatic detection and real-time speaker labels.

    For buildersTry it on live audio where speaker labels need to appear during the session, and check whether its transcript-focused turn segmentation fits the interaction.

  3. AssemblyAI

    AssemblyAI-STT-3.7-Realtime-Agent

    AssemblyAI released AssemblyAI-STT-3.7-Realtime-Agent for voice-agent conversations. It supports 32 languages with automatic detection and is designed to focus transcription on the person speaking to the agent.

    For buildersEvaluate it on calls with background speech and short numeric responses, checking transcripts against consented test audio before deployment.

  4. AssemblyAI

    Universal-3.6 Pro Realtime

    AssemblyAI released Universal-3.6 Pro Realtime, an upgrade to Universal-3.5 Pro for streaming speech recognition. It supports 32 languages with automatic detection and is designed to handle short responses, multiple speakers, accents, and noisy telephony audio at the previous version’s latency.

    For buildersTest it on a representative set of noisy calls with short confirmations, names, and code-switching, and measure transcription errors before connecting its output to automated actions.