Audio edition of a weekly newsletter: 52 issues cost $9.88 on Sume

A year of audio editions for a weekly newsletter costs $9.88 on Sume TTS 1.0 if each issue is 4,000 characters. At $0.0475 per 1,000 characters an issue is $0.19 exactly, 52 issues are 208,000 characters, and nothing rounds up because 4,000 characters land on a whole cent (read 2026-10-06). The same 208,000 characters list at $4.58 on MAI-Voice-2.1 ($22 per million) and $3.12 on MAI-Voice-2.1-Flash ($15 per million), and neither is available through Sume.

So Sume is the pricier line on raw characters, by about $5 to $7 a year at this volume. What the premium buys is the same wallet, ledger and API that you use for the video and image side of a newsletter, plus a refund on a failed job. If audio is the only thing you generate and you have a direct vendor account, compare honestly. If you are already generating other media on Sume, the cost of a second account is probably larger than the five dollars.

Where the numbers come from

Sume bills the provider list price times 1.25, rounded up to the cent per job. For TTS that is $0.0475 per 1,000 characters, so a 2,000-character issue is $0.095 and bills $0.10, and a 10,000-character issue is $0.475 and bills $0.48. The MAI columns are straight multiplication of the vendor's published per-million figures, with no rounding, as a reader would compute it from the vendor's page.

Those vendor figures are list prices from a tracker page and the launch announcement, and they say nothing about quotas, regions or volume deals. A real quote could differ. Use them for a first comparison, then check Microsoft's own pricing page before you decide.

• Issue length drives the cost, so trim the reading copy, not the web copy: links, footers and image captions do not need to be read aloud.
• One job per issue gives one file per issue, which is what a podcast feed wants.
• Do not generate the audio until the text is final: a late edit means a second $0.19.
• A failed job is refunded, so a retry after a failure costs nothing.

A yearly estimate you can run

The script prices a year for any issue length and shows the Sume bill beside the two vendor list prices.

def sume_cents(chars: int) -> int:
    return -(-chars * 475 // 100_000)


def year(chars: int, issues: int = 52) -> dict:
    return {
        "sume": round(sume_cents(chars) * issues / 100, 2),
        "mai_2_1": round(chars * issues * 22 / 1e6, 2),
        "mai_flash": round(chars * issues * 15 / 1e6, 2),
    }


for n in (2000, 4000, 6000, 10000):
    print(n, year(n))

Reading copy, not web copy

Most newsletters carry text that should not be spoken: a header with a date and a logo, a row of share links, a sponsor block and a footer with an address. If those are 700 of the 4,000 characters, you are paying for 18 percent of the audio to read things nobody wants to hear. Build a reading copy from the issue by dropping those parts, and price that copy. At 3,300 characters an issue bills $0.16, which is $8.32 for the year, a saving of $1.56.

Numbers and URLs also read badly. Spell out the figures you want said and remove web addresses, then replace them with a short phrase such as 'link in the show notes'. The audio is shorter, cheaper and easier on the ear, and the text copy still has everything. Listeners subscribe to a feed, so give each file a title and date, and publish it with the written issue.

Adding a cover per issue

If every audio edition also gets a square cover, a low-tier Ideogram 4.5 image is $0.0375, which adds about $1.95 for 52 issues, and a medium-tier image is $0.075, about $3.90. The two together at the low tier put an issue at about $0.23. Read the live rates in the pricing page and the vendor tracker.