2.1 million TTS characters a month on Sume: $99.75 in 105 requests

Two million one hundred thousand characters of text-to-speech cost $99.75 on Sume: 2,100 thousand-character blocks x $0.0475. The catalog caps one request at 20,000 characters, which prices at $0.95, so the month is 105 full requests (2,100,000 / 20,000) or more if you split by script.

Spaces and punctuation count as characters, per the catalog description. Budget on the length of the string you send, not on the word count.

Sizing the requests

The table prices four request sizes against the 2.1 million character month. All values use the catalog rate of $0.0475 per 1,000 characters, read 2026-10-09.

The cent rounding

The catalog says the submit-time reservation stores the estimated Sume amount in USD micros and rounds up to cents for balance compatibility. For a 1,000-character request the exact price is 4.75 cents and the catalog estimate is 5 cents. Treat the cent figure as the hold and the per-1,000 rate as the pricing basis the catalog states.

One more detail from the catalog: the applied estimate adds an agent fee line (550 basis points in the price book), and the applied estimate for 1,000 characters is 6 cents rather than 5. If your pipeline runs through an agent, use the applied figures.

Concurrency, not price, sets the pace

At 105 requests a month the money is small and the shape is the issue. Each request is a paid generation job, so it takes a seat in admission: Free processes 1 at a time with room for 6 accepted jobs, Pro 4 and 24, Startup 8 and 48. A script that submits all 105 on Pro at once will get 429 queue_full after the 24th. Submit in waves and keep an Idempotency-Key per script so a retry does not pay twice.

Budget lines worth adding

Add a retry allowance. A 10 percent re-record rate on 2.1 million characters is 210,000 more characters, $9.975, which brings the month to $109.725. Add the hold: with 4 jobs processing on Pro and a full 20,000-character request at $0.95 each, about $3.80 sits reserved at any moment, which is the minimum balance buffer for steady operation. Voice quality problems usually come from the script, not the synthesis, so fix text before regenerating audio; a regenerated clip bills in full.

Splitting a script

Long scripts are cheaper to fix when they are split. A 20,000-character request that needs one sentence changed bills $0.95 again if you regenerate the whole request, while a 1,000-character paragraph bills 4.75 cents. Splitting by paragraph or scene keeps regeneration small and also spreads work across the concurrency seats of your plan. The cost is more jobs and more stitching, so choose a size that matches how often your scripts change after review.