Model comparison
Compare MAI-Transcribe-2 and Universal-3.5 Pro using the same provider-sourced speech recognition & transcription rubric. No mystery score and no invented benchmark ranking.
Facts checked September 4, 2026
Set your usage. Your estimate updates as you type.
Uses the recording length in hours: 30 minutes = 0.5 hours. Extra features and minimum charges may change the bill.
Estimates exclude taxes, tools, cache storage/writes, free allowances and custom discounts. Image estimates cover output only, not prompt or reference-image charges. Quality modes differ by model. Unlisted settings are not treated as free.
| Model | Access | Estimated total (USD) |
|---|---|---|
| MAI-Transcribe-2Microsoft | Microsoft | No reviewed rate |
| Universal-3.5 ProAssemblyAI | AssemblyAI | No reviewed rate |
Quick take
Microsoft's fast batch transcription model for long recordings, speaker labels, word timing, keyword biasing, and clean or verbatim transcripts.
Microsoft's one-hour-in-about-ten-seconds figure is a provider claim, not a guaranteed SLA. Benchmark your own audio and confirm the post-preview price, deployment region, quotas, and retention configuration.
AssemblyAI's accuracy-focused transcription model for files and live audio, with code-switching, speaker labels, timestamps, and contextual prompting.
Async and realtime routes have different prices, and retention depends on endpoint and account controls. Verify optional feature charges, EU routing, training opt-out, BAA status, and zero-retention eligibility.
Compare the published facts
Values use each provider's own published units and limits. A blank means the provider did not publish a directly comparable value in the sources reviewed.
| Speech recognition & transcription | MAI-Transcribe-2 | Universal-3.5 Pro |
|---|---|---|
| Transcription priceCurrent provider price per audio hour or minute for the listed processing route. | $0.10 / audio hour limited-time rate | $0.21/hour async · $0.45/hour streaming |
| Live or batchWhether the model handles realtime streams, uploaded recordings, or both. | Fast batch and long-form transcription | Uploaded files and realtime streaming |
| LanguagesProvider-published language coverage, separating trained or advertised coverage from specifically verified languages where needed. | 60 languages | 18 languages at launch with native code-switching |
| Speaker labelsWhether the model identifies who spoke and any published speaker limit. | Yes | Yes |
| TimestampsAvailable word-, segment-, or utterance-level timing information. | Word-level timestamps | Word-level timestamps |
| Vocabulary controlKeyword boosting, custom spelling, context, prompting, or other ways to improve domain terms. | Keyword biasing, clean/verbatim output, automatic language detection | Contextual prompting and keyterm prompting |
| Where to use itDirect API, cloud catalog, application, or regional endpoint documented by the provider. | Microsoft Foundry / Azure | AssemblyAI US and EU APIs |
How to choose
Start with the job you need to complete, then validate cost, access, and policy details on your exact provider route.
Microsoft says prompts, outputs, embeddings, and training data submitted to Foundry Models are not available to model providers and are not used to train foundation models without permission. Retention and abuse-monitoring details depend on the deployed service.
AssemblyAI's retention and model-training behavior depends on endpoint, contract, BAA status, EU processing, and opt-out settings. Its docs describe zero-data-retention options for eligible realtime use; verify the exact account configuration.
Provider and API links
Frequently asked questions
MAI-Transcribe-2: Microsoft's fast batch transcription model for long recordings, speaker labels, word timing, keyword biasing, and clean or verbatim transcripts. Universal-3.5 Pro: AssemblyAI's accuracy-focused transcription model for files and live audio, with code-switching, speaker labels, timestamps, and contextual prompting.
Consider MAI-Transcribe-2 when your priority is High-volume recorded calls and meetings. Consider Universal-3.5 Pro when your priority is Meetings and calls with language switching. Test both with your own data and provider route before committing.
No. This comparison aligns provider-published facts for the Speech recognition & transcription category. It does not claim a universal winner or combine incompatible third-party benchmark scores.
