Skip to main content
AssemblyAI regularly releases new speech models with better accuracy, lower latency, and broader language support. As newer models become generally available, older models are deprecated and eventually retired. This page covers AssemblyAI speech models only: speech-to-text and text-to-speech. It explains the model lifecycle, what happens when a model is retired, and the current status of each speech model.

Model lifecycle

Every AssemblyAI speech model moves through the following stages.

Model status

Pre-recorded models

Default model change: For new accounts created on or after November 17, 2026, requests that don’t set speech_models use assemblyai-stt-3-7. Starting December 8, 2026, existing accounts that don’t set speech_models and currently use universal-3-5-pro or universal-3-pro by default will automatically use assemblyai-stt-3-7. This is a change to the default model, not the retirement of universal-3-5-pro or universal-3-pro. Existing accounts that use universal-2 by default aren’t upgraded automatically. See the assemblyai-stt-3-7 migration guide.

Real-time models

Sync models

Text-to-speech models

Retired models

Retired models are listed below, with the most recent retirements first.

Model deprecation and retirement

Notification

AssemblyAI gives you at least 90 days’ notice before retiring a model.

What happens when a model is retired

When a model is retired, AssemblyAI routes any request that specifies it to its replacement model if available. We recommend that you evaluate the replacement model and test it on your own audio before the retirement date.

Auditing model usage

To find out whether you’re using a deprecated model, review your usage in the AssemblyAI dashboard:
  1. Open the Usage page.
  2. Break down your usage by model to see which models your requests use.