MAI voice and transcribe prices come from one report: verify first
The $22, $15 and $0.54 MAI figures all come from one Unite.AI article. What it says, which numbers have a stated end date, and what to confirm before budgeting.

Short answer
If you are about to budget for MAI-Voice-2.1 or MAI-Transcribe-2-Streaming, know where the prices come from: in this series they all trace to a single secondary article, and the article itself carries a disclosure that matters. Confirm each figure on Microsoft's own pricing page before you put it in a contract or a forecast.
Everything below is from the Unite.AI article on the October 1, 2026 launch (read 2026-10-07). It labels its author, Jonas Reeve, as an AI-generated research agent, and says articles by that author are AI-generated and reviewed by Unite.AI's editorial team.
The article states these prices and terms. The last column is what to check, not a claim that anything is wrong.
| Item | What the article says | Check on the vendor page |
|---|---|---|
| MAI-Voice-2.1 | $22 per 1M characters | Currency, billing unit, any minimum |
| MAI-Voice-2.1-Flash | $15 per 1M characters | Same, plus the 45-second audio limit |
| MAI-Transcribe-2-Streaming | $0.54 per hour of audio, introductory, through the end of the year | What the price becomes after the intro period |
| Availability | Foundry, MAI Playground, Vercel, Azure Voice Live for all three; OpenRouter for the two voice models; LiveKit listed as coming soon | Which channel you will buy through, since resale channels can price differently |
What the numbers leave open
Three points stand out. The transcription price is introductory, and the article does not give the later price, so a twelve-month budget cannot be built from it. The article also says the voice prices are for specific models, and that availability differs by channel, so "the MAI price" depends on where you buy. Finally, the benchmark and latency numbers in the same article are Microsoft's or Artificial Analysis's claims, relayed. Quote them with attribution.
What to do instead
Prices for a product you can use today live on a catalog you can query. Sume publishes live rates at GET /v1/catalog, and the docs list the public rate on each surface: $0.01 per audio minute for speech-to-text, $0.01 per job for audio detach and for timeline audio joins, and $0.20 per standalone caption job of up to 60 seconds. Text-to-speech is metered per character, and the TTS Router publishes each model's list price per character at GET /v1/tts-router/models.
The practical habit is the same for any vendor. Put the price, the date you read it and the URL you read it from in the first row of your cost sheet. When a number changes, the sheet shows which assumption moved.
A four-line verification routine
- Open the vendor's pricing page and copy the figure with the read date.
- Note whether the price is introductory, preview or list.
- Multiply by your own volume, not by the volume in a launch example.
- Re-check on the first of each month while a launch price is in effect.
A budget line that survives a price change
Put the price in a single cell of your budget, label it with the source and date, and compute every other cell from it. When the vendor page changes, you edit one number and the sheet follows.
For anything above a pilot, confirm the price with the vendor's own pricing page or a quote, since this post relied on one secondary report.
- One price cell, with source and date.
- Everything else is a formula.
- Confirm with the vendor before committing spend.
Sources
Related posts
More in Pricing
- One 16:9 clip cropped to 9:16 and 1:1: $0.04, not 3 generations
A 15-second 720p Wan clip is $1.875. Cropping it to 9:16 and 1:1 with two $0.02 video-filter jobs adds $0.04, against $3.75 to generate both shapes again.
- What does one hour of AI narration cost on Sume, MAI and ElevenLabs?
One hour of finished narration is about 45,000 characters: $2.16 on Sume in three jobs, against $0.99 on MAI-Voice-2.1. Rates read 2026-10-07.
- One music bed for 50 holiday ads: $0.125 total, $0.0025 each
A Lyria music track is a flat $0.125 on Sume. One bed reused across 50 ads is $0.0025 per ad; a fresh track for every ad costs $6.25.
- What is a Q4 video budget for a one-person shop, week by week?
Ten weekly jobs from Halloween to New Year, built from 720p Omni clips, Ideogram images, voice, music and captions, come to $9.675 on Sume.
Written by Sume