AssemblyAI launches promptable speech model Universal-3 Pro
In short: AssemblyAI has introduced Universal-3 Pro, a specialized speech-to-text model that allows developers to provide instructions directly to the transcription process. It supports verbatim and clean transcription, custom audio event tagging, and handles code-switching between six languages.
This summary was generated automatically by AI from AssemblyAI's publication. It is our own text, not a copy of the original — facts, figures and quotes belong to the source, linked above and below.
What changed?
- 1New promptable architecture for instruction-guided transcription.
- 2Introduced keyterm prompting for a reported 45% accuracy boost on domain-specific vocabulary.
- 3Native support for six languages with automatic code-switching.
- 4Pricing set at $0.21 per hour with no rate limits.
| Parameter | Before | Now |
|---|---|---|
| Price | Not specified | $0.21/hour |
| Languages | Not specified | 6 natively (English, Spanish, German, French, Portuguese, Italian) plus auto-routing to 99 |
| Capabilities | Not specified | Promptable transcription, verbatim/clean control, custom audio tagging, speaker diarization, code-switching |
Why it matters
This model removes the need for complex, error-prone post-processing pipelines by enabling instructions like 'include all disfluencies' or 'label speakers by role' at the transcription source. It is designed to be more reliable and cost-effective than general-purpose multimodal foundation models for voice-heavy production applications.
What it means for AI agents and contact centers
When testing this against existing STT pipelines, compare the accuracy of domain-specific terminology and speaker identification against your current vendors. Evaluate if the built-in prompting for verbatim records reduces your reliance on external LLM layers for cleaning transcripts, and measure how the $0.21/hour cost-performance compares to your current transcription load.
🧪 Worth Testing
It offers potential reductions in latency and overhead by moving formatting and entity extraction logic into the model layer.
Sources
- AssemblyAIOfficialPrimary sourceOriginal article →„Introducing Universal-3 Pro: A new class of speech language model optimized for Voice AI“
- Published by source
- —
- Found by our system
- 29 Oct 2026, 21:55
- Summary generated
- 29 Oct 2026, 22:26
This article was written by AI from the original source. Facts, numbers and prices come from the source; missing values are marked “Not specified”. Legal notice, copyright and privacy