Native code-switching across 18 languages
The launch article says the model handles code-switched speech natively across 18 languages, so transcripts can follow the language actually spoken without a separate language pass.
Universal-3.5 Pro
Universal-3.5 Pro is AssemblyAI’s async speech-to-text model for recorded audio and video. It adds native code-switching, speaker diarization, and contextual prompting, with a base price of $0.21 per hour of audio submitted.
Overview
Universal-3.5 Pro is AssemblyAI’s flagship asynchronous speech-to-text model for recorded audio and video. It is part of the Pre-recorded Speech-to-Text API and is positioned as the highest-accuracy model for supported languages and difficult real-world audio.
The launch article highlights three main capabilities: native code-switching across 18 languages, improved speaker diarization, and contextual prompting. The pricing page lists the base rate at $0.21 per hour of audio submitted, with optional add-ons for capabilities such as diarization, medical mode, keyterms prompting, and general prompting.
In practice, the model is aimed at transcription workloads where speech is messy, multilingual, or speaker-heavy. That includes meetings, contact center recordings, captions, searchable archives, and other recorded workflows where accuracy and speaker attribution matter more than live latency.
Core capabilities
Improved speaker diarization for overlapping speech
AssemblyAI describes this release as its most accurate speaker diarization yet, with the model designed to keep speaker attribution aligned with the transcript even in crosstalk and noisy conversations.
Contextual prompting
Contextual prompting lets you provide domain knowledge or prior context to influence the transcription output, which can improve accuracy on specialized terms or ambiguous audio.
Keyterms prompting for custom vocabulary
The pricing reference lists keyterms prompting for Universal-3.5 Pro with support for up to 1,000 terms, giving teams a way to bias recognition toward project-specific vocabulary.
Pre-recorded transcription model
Universal-3.5 Pro is the async pre-recorded model in AssemblyAI’s speech-to-text lineup and is priced at $0.21 per hour of audio submitted.
Add-on support for specialized workflows
The pricing page shows optional add-ons for this model, including medical mode and speaker diarization, which can be combined with the base transcription rate.
Common use cases
Multilingual recorded conversations
Use Universal-3.5 Pro when a recording contains multiple languages in the same conversation and you want the transcript to preserve the language spoken at each point.
Meetings with overlapping speakers
Use the model for meeting notes or internal discussions where short turns, interruptions, and overlapping voices make speaker attribution difficult.
Contact center recordings
Use it for contact center calls or other customer interactions where accurate speaker labeling affects downstream analysis and QA.
Domain-specific transcription
Use contextual prompting or keyterms prompting when transcribing specialized discussions that include project names, domain terms, or other vocabulary that may be misheard.
Batch transcription of recorded media
Use the async API for archives, captions, or dataset preparation when the audio is already recorded and you can trade latency for transcription quality.
Pros and Cons
Pros
- Supports native code-switching across 18 languages, which is useful when conversations move between languages mid-sentence or mid-call.
- Adds contextual prompting and keyterms prompting to help guide transcription toward domain-specific vocabulary.
- Improves speaker diarization for conversations with crosstalk, interruptions, and short turns.
- Priced separately from real-time models, with a clear async rate in the pricing reference.
- Fits a broad set of recorded-audio workflows, including meetings, contact center audio, media, and archives.
Cons
- It is an async model, so it is not the right fit for live transcription use cases that need streaming output.
- Several advanced capabilities are billed as add-ons rather than being included in the base transcription rate.
- The source material does not provide a full list of supported languages, output formats, or setup details for the model itself.
FAQ
What kind of transcription workflow is Universal-3.5 Pro for?
Universal-3.5 Pro is AssemblyAI’s async pre-recorded speech-to-text model. It is designed for recorded audio and video rather than live streaming transcription.
What language support does it offer?
The launch article says it supports native code-switching across 18 languages. The product pages identify it as the highest-accuracy pre-recorded model for supported languages.
Can I guide the model with extra context?
The launch article highlights contextual prompting, and the pricing reference shows general prompting is available as a paid async add-on for Universal-3.5 Pro.
Does it support speaker diarization?
The pricing page lists async speaker diarization for Universal-3.5 Pro as a paid add-on, with standard and experimental variants. The launch article positions diarization as one of the model’s core improvements.
How is Universal-3.5 Pro priced?
The pricing page lists Universal-3.5 Pro at $0.21 per hour for pre-recorded audio submitted. Add-ons such as diarization, medical mode, keyterms prompting, and general prompting are priced separately.
Quick Facts
- Category
- Pre-recorded speech-to-text API
- Platform
- AssemblyAI Cloud and self-hosted Voice AI infrastructure
- Primary users
- Developers building transcription workflows for recorded audio and video
- Base price
- $0.21/hr for pre-recorded audio submitted
- Model API value
- universal-3-pro
- Source domain
- assemblyai.com
Universal-3.5 Pro Alternativen
QuickQuill
QuickQuill is a macOS dictation and transcription app that runs locally on the device. It helps users record meetings, transcribe audio, generate summaries, and export notes without using a cloud service.
Speech to Text Converter
Speech to Text Converter is a browser-based transcription tool for live dictation and uploaded audio or video files. It offers a free tier for short tasks and a Pro plan for unlimited transcription, AI summaries, translation, speaker identification, and advanced exports.
Realtime and audio
An OpenAI API guide for choosing the right speech architecture for live audio, translation, transcription, speech generation, and audio-capable chat. It helps developers map each speech application to the appropriate session type, endpoint, and connection method.
Pewbeam
Pewbeam is a church presentation app that listens to sermons, detects Bible verse references in real time, and displays the matching passage on screen. It is built for pastors, projection teams, and church media volunteers who want to reduce manual slide control during live services.
Dictato
Dictato is a Mac dictation app that transcribes speech into text in any app using an on-device, offline workflow. It supports multiple transcription engines, optional cleanup and translation, and a one-time purchase license.
Voicenotes
Voicenotes is an AI note-taking and meeting recording app that transcribes conversations, generates summaries and action items, and makes past recordings searchable with Ask AI. It also supports voice dictation, imported audio on Pro, and multilingual transcription.