Narrate a 40,000-character course: MAI-Voice, Eleven v4 and Sume cost

The numbers

For 40,000 characters, MAI-Voice-2.1-Flash costs $0.60, MAI-Voice-2.1 and the promotional Eleven v4 rate cost $0.88 each, and Sume TTS costs $1.90. Divide each per-million price by 25 to get the 40,000-character cost, since 40,000 is 1/25 of a million.

Sume is the most expensive of these. The reason to pay it is the shape of the service, not the rate, and the rest of this post says what that means.

Price table

Prices are copied from each vendor's own page, and the totals are our arithmetic. Eleven v4's low rate is a launch price; the same page lists a higher regular price.

What the lower rates come with

Microsoft's announcement lists MAI-Voice-2.1 at $22 and MAI-Voice-2.1-Flash at $15 per million characters, and its voice documentation marks both as public preview with no SLA. You also need an Azure setup. ElevenLabs' API pricing page shows v4 at $22 per million against a regular $80, and our earlier check found the promotion ending October 12.

Sume's rate is $0.0475 per 1,000 characters, which is $47.50 per million. In return you get one HTTP API with job ids, idempotency keys, signed webhooks and a pay-as-you-go balance, with no per-vendor account to open. If you narrate 40,000 characters a month, the gap with the cheapest option is $1.30 a month. If you narrate 40 million, it is $1,300, and the choice is different.

How to decide

Volume decides. At course scale, a dollar of difference rarely matters; the time to wire up a vendor does. At catalogue scale, the rate dominates and a direct vendor contract is worth negotiating.

• Under a million characters a month: pick on workflow, not rate.
• Over ten million: price the direct vendor and ask for a contract.
• Any preview service: keep a fallback route in case terms change.
• Re-check every number on the day you buy; promos end.

A worked example: eight modules

Split the 40,000 characters into eight modules of 5,000 characters. One module costs $0.075 on MAI-Voice-2.1-Flash, $0.11 on MAI-Voice-2.1 or promotional Eleven v4, and $0.2375 on Sume. Per-module jobs matter because courses change: if an instructor rewrites two modules after launch, you pay for 10,000 characters of re-takes, not 40,000.

Budget for re-takes up front. If 20 percent of the script is regenerated once, that is 8,000 extra characters, which adds $0.12 on Flash, $0.176 on the $22 services and $0.38 on Sume. The extra is small, but it is the part that grows when a script goes through review.