Sume TTS is 2.16x MAI-Voice-2.1 per character: where it still fits

Sume text to speech costs 2.16 times MAI-Voice-2.1 and 3.17 times MAI-Voice-2.1-Flash per character, and the ratio does not change with volume, so there is no character count at which Sume becomes the cheaper option. Sume lists $0.0475 per 1,000 characters ($47.50 per million), the MAI-Voice-2.1 page lists $22 per million characters, and the Flash entry lists $15 per million. The honest question is what the Sume premium pays for, and whether your job needs it.

The ratio and the gap at three volumes

The ratios are plain division: 47.50 / 22 = 2.159 and 47.50 / 15 = 3.167. The gap column is Sume minus Microsoft, so a positive number is what you would pay extra. The volumes are arbitrary on purpose: 88,000 characters is a long audiobook chapter set, 640,000 a small catalog, 2,750,000 a large one.

What is different about a Sume job

A Sume TTS request is a job: you submit it, the default mode is async, and you poll GET /v1/jobs/:id/status and read GET /v1/jobs/:id/result, or pass a webhook. The finished audio is a media.sume.com artifact rather than a response body you have to store yourself, and the job records the engine, voice, language, output format and synthesis settings so the next line can be made to match (Jobs and results). Those are workflow properties. They do not change the per-character price, and Sume does not list any MAI model, so you cannot buy MAI-Voice through Sume.

The catalog caps one TTS request at 20,000 characters, which is $0.95 at the Sume rate; the same 20,000 characters is $0.44 at the MAI-Voice-2.1 rate and $0.30 at the Flash rate. The Microsoft model page does not state a request cap in the fields I read, so I make no claim about theirs.

When the premium is not the deciding number

If the spend is dominated by video generation, a voiceover is a small line. A 25-second ad read of 400 characters costs $0.019 on Sume against $0.0088 on MAI-Voice-2.1; the difference is $0.0102, less than the 1 cent per-job floor on its own. If you generate 100,000 such lines a month, the difference is $1,020.00 for the month, and that is the case where moving the voice step to a cheaper provider and importing the audio is worth building. Microsoft's page labels MAI-Voice-2.1 for audiobooks, content creation and voice-over and Flash for latency-sensitive uses such as call-center agents, voice assistants and IVR, so match the model to the job before comparing prices.

Check the arithmetic on your own volume

Take your last month's character count N. Sume is N x $0.0000475, MAI-Voice-2.1 is N x $0.000022 and Flash is N x $0.000015. For N = 1,250,000 that is $59.38, $27.50 and $18.75, so the Sume premium is $31.88 over 2.1 and $40.63 over Flash. Add one cent for every job under 211 characters on Sume if your lines are short. Then ask whether the workflow value, such as durable job results, polling or webhooks, and the Timeline hand-off, is worth that number to you.