Sume TTS vs MAI-Voice-2.1 at 50k to 5M characters a month

At 1 million characters a month, Sume TTS 1.0 bills $47.50, Microsoft's MAI-Voice-2.1 lists $22 and MAI-Voice-2.1-Flash lists $15 (read 2026-10-06). On raw character price MAI-Voice is cheaper, by a factor of 2.2 for the standard model and 3.2 for Flash. MAI-Voice is not a Sume model, so you cannot choose it through Sume.

That is the honest starting point for a buyer who types "MAI-Voice cost vs Sume TTS". The next question is what the extra $25.50 a month at 1M characters buys, and for most teams the answer is the surrounding workflow rather than the voice: one key, one wallet, one job envelope, and TTS that sits next to video, captions and the timeline render.

How the Sume column is built

Sume bills $0.0475 per 1,000 characters and caps one job at 20,000 characters, which is $0.95. A volume of 1,000,000 characters is 50 full jobs, so $47.50. A volume of 250,000 is 12 full jobs and one 10,000-character job at $0.48, so $11.88. The MAI columns are volume times the list price per million characters, with no rounding because the vendor page does not state a per-request rule.

The MAI figures come from a model-release tracker that cites the vendor launch of 2026-10-01, which also lists 23 languages and a Flash variant that the vendor claims reaches 150 ms end to end. Treat the Flash latency as a vendor claim until you have measured it.

• Sume TTS jobs are asynchronous: submit, poll GET /v1/jobs/:id/status, then read the result.
• Sume's TTS Router docs list streaming TTS as a non-goal of its first release, so a 150 ms conversational loop is not what Sume TTS is for.
• Each job rounds up to the next cent, so many short jobs cost slightly more than the table, which packs full 20,000-character jobs.
• The gap between the columns is the price of the workflow, so decide whether you use it.

Compute your month

The function below splits a monthly volume into 20,000-character jobs and prints the Sume bill next to the two MAI list prices.

RATE = 47.5 / 1_000_000   # dollars per character on Sume TTS 1.0
CAP = 20_000


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


def sume_dollars(chars: int) -> float:
    full, rest = divmod(chars, CAP)
    cents = full * job_cents(CAP) + (job_cents(rest) if rest else 0)
    return cents / 100


for chars in (50_000, 250_000, 1_000_000, 5_000_000):
    mai, flash = chars * 22 / 1e6, chars * 15 / 1e6
    print(chars, sume_dollars(chars), round(mai, 2), round(flash, 2))

What does not change with volume

Sume's rate is flat across these four volumes. There is no tier that lowers the per-character price at 5 million characters, so the ratio to the MAI list prices stays at 2.2 and 3.2 all the way up the table. That makes the decision a stable one: if the voice is the only line item, volume does not close the gap, and if it is one of several, volume does not widen it.

One more check before you compare: confirm the vendor's price on its own page on the day you buy, since the Flash price and the transcription intro price are both new. A list price you read last week is a guess this week.

A scenario with real numbers

Take a team that ships 40 explainer videos a month, each with a 1,500-character voice-over. That is 60,000 characters. On Sume each voice-over is its own job at $0.0475 per 1,000 characters, 7.125 cents, which bills as $0.08 after rounding up to the cent, so $3.20 for the 40; MAI-Voice-2.1 lists $1.32 and Flash $0.90 for the same volume. The voice line is a rounding error next to the video line, which is $50.00 for 40 ten-second Wan 3.0 clips at 720p. At that scale the price of the voice should not decide the vendor; fit, quality and operations should.

Who should pay the difference

If speech is the only thing you buy and your volume is in the millions of characters, the list prices point at MAI-Voice, and a Sume-priced voice is hard to justify. If speech is one step in a pipeline that also renders video, burns captions and stitches a timeline, a second vendor means a second key, a second balance and a second usage export to reconcile, and you can decide whether that is worth $25 a month to you. Read the Sume API rate card for the live TTS rate, and the tracker for the vendor's launch details.