What $137 buys: MAI-Voice, Gemini TTS, Sume TTS and STT in units

A $137 budget buys 2,884,211 characters of Sume TTS, 6,227,273 characters of MAI-Voice-2.1, 9,133,333 characters of MAI-Voice-2.1-Flash, 169.1 hours of Gemini 3.8 Flash TTS audio output, 228.3 hours of Sume STT, 1,370 hours of MAI-Transcribe-2 batch and 253.7 hours of the streaming model. The units differ, so compare within a row family, not across it.

Each budget as units

Each row is $137 divided by the vendor's listed unit price, read 2026-10-08. Gemini's audio figure is the output charge only (25 audio tokens a second: 90,000 tokens an hour x $9.00 per million = $0.81, doubling to $1.62 from January 1, 2027) and excludes its text-input tokens. Sume TTS ignores the 1-cent job floor, so a budget spent on short lines buys fewer characters.

Why the units are not interchangeable

Characters become audio at your speaking rate. At an assumed 900 characters a minute, 2,884,211 Sume characters is 3,204.7 minutes (53.4 hours) of speech, 6,227,273 MAI-Voice-2.1 characters is 6,919.2 minutes (115.3 hours) and 9,133,333 Flash characters is 10,148.1 minutes (169.1 hours). At exactly 900 characters a minute the Flash figure equals the Gemini audio row, because 54,000 characters an hour x $15 per million is $0.81; that is the assumption doing the work, and Gemini's row does not depend on it.

Speech-to-text is metered in audio time on every service in the table, so the STT rows compare directly: Sume's $0.60 an hour buys 228.3 hours, Microsoft's streaming model 253.7, its batch model 1,370.

The pattern

Per hour of speech, the two Microsoft voices and the Gemini output charge are all cheaper than Sume TTS at normal speaking rates, and Sume STT is dearer than both Microsoft models. What Sume adds is the job-and-Timeline workflow around the audio, so spend $137 on Sume speech only if that workflow is what you are buying.

Do the same for your own budget

Replace 137 with your monthly figure and the table scales linearly. Because every price here is linear, there is no volume discount to find in the numbers I read: a budget of $1,370 buys ten times each quantity. The exceptions are the rules that are not linear: Sume's 1-cent job minimum, Gemini's doubling on January 1, 2027, and the Microsoft introductory transcription prices.

Remember that these are list rates read on 2026-10-08. They exclude taxes, free tiers, committed-use discounts and anything vendors apply to your account, and the Gemini figures assume 25 audio tokens a second, which is the rate Google's page states.