TTS per 1,000 characters: Voxtral $0.016, MAI $0.022, Sume $0.0475

On list price alone, Sume TTS at $0.0475 per 1,000 characters is about three times Mistral's Voxtral TTS API at $0.016 and about twice Microsoft's MAI-Voice-2.1 at $0.022 per 1,000 characters. MAI-Voice-2.1-Flash is $0.015. For a 10,000-character script that is $0.16, $0.22, $0.15 and $0.475 respectively. The difference buys a job envelope, hosted artifacts and timings; whether it is worth it depends on how much of the pipeline you would otherwise build.

The numbers

Vendor prices are as stated on the pages read on 2026-10-04. Sume's price is Cartesia's list credit rate times Sume's 1.25 margin, per the TTS router billing notes; confirm it in the live catalog.

What the Sume price includes

Each Sume TTS call is a job with status, result and cancel URLs, an Idempotency-Key, signed webhooks and a wallet reservation. The result is a hosted artifact on media.sume.com, which other Sume surfaces read directly. With one flag the job also returns word timings, and with sentence segmentation it returns gapless per-sentence slices; the API reference has the schema.

Those are the pieces you would otherwise write around a raw speech API: storage, retries, timing extraction, slicing. If you already run that machinery, the raw API is cheaper per character. If you do not, count the engineering time before you count the characters.

Where the price stops mattering

For short videos the totals are small whichever you pick. A 900-character one-minute script costs about 4.3 cents on Sume at list. The big costs in a video pipeline are usually the video and the retakes, not the voice. Still, long-form work scales linearly: a 20,000-character job, the Sume maximum, is $0.95 at list.

Ways to spend less on any service

Fix the script before you voice it. Re-record single lines rather than whole scripts. Join takes with timeline audio instead of regenerating, which is a $0.01 job with no re-synthesis. And check the live number in GET /v1/catalog, because list prices move.