#

Key takeaways

• Microsoft AI released MAI-Transcribe-2.

• The introductory price is $0.10 per audio hour.

• The prior model cost $0.36 per audio hour.

• The model adds speaker diarization, configurable transcription styles, and word-level timestamps.

• The source reports no Morocco rollout.

What the source says

VentureBeat reported on 2026-09-03 that Microsoft AI released MAI-Transcribe-2. The source says the model is priced at an introductory $0.10 per audio hour. It also says the prior model cost $0.36 per audio hour.

The reported update is not only about price. The source says MAI-Transcribe-2 adds speaker diarization, configurable transcription styles, and word-level timestamps. These features matter because they can change how teams review audio and assign spoken segments.

Claimed performance and benchmark context

The source says Microsoft claims MAI-Transcribe-2 leads the FLEURS benchmark across 60 languages. It also says the model has a 5.2% average word-error rate. Those are vendor claims in the supplied material, so they should be treated as claims rather than independent verification.

That matters for buyers and editors. Benchmark leadership can be useful, but it does not replace testing on real audio. The source does not provide independent validation, so the practical value still depends on local evaluation.

What the new features change

Speaker diarization separates who spoke when. That can help with calls, meetings, and interviews. Word-level timestamps can also make review and editing easier.

Configurable transcription styles suggest that users may be able to shape output for different needs. The source does not explain the available styles, so any deeper interpretation would be an assumption. Still, the feature points to more control over transcript format.

The lower introductory price may also affect adoption decisions. A cheaper per-hour rate can reduce the cost of testing and early deployment. The source does not say how long the introductory price will last.

Operational considerations

The source points to a practical question: testing language, accent, privacy, and correction quality locally before relying on the model. That is a sensible caution because speech systems can behave differently across audio conditions. The source does not provide details on data handling, retention, or deployment settings.

Correction quality is especially important. A transcript can look polished while still missing names, terms, or speaker changes. The source does not say how MAI-Transcribe-2 handles those edge cases, so teams would need their own checks.

Morocco relevance

The source reports no Morocco-specific rollout or local program. For readers in Morocco, the global lesson is simple: test speech models on your own audio before adoption. That is an assumption-based operational step, not a Morocco-specific claim.

Why this release matters

MAI-Transcribe-2 combines three practical improvements in one release: lower pricing, speaker separation, and timestamped output. That mix can make speech recognition easier to use in workflows that depend on review and editing. The source frames the release as a product update with measurable pricing and feature changes.

The benchmark claim also gives the release a performance narrative. But the source does not show how the model performs on every accent, recording quality, or domain. That gap is why local testing remains important.

What to watch next

The source does not report a rollout plan, regional availability, or enterprise terms. It also does not describe any migration path from the prior model. Those missing details limit how far readers can go from the announcement alone.

For now, the safest reading is straightforward. Microsoft AI has introduced a cheaper speech-recognition model with added transcript controls. The real test will be how it performs on the audio that matters to each user.