3,000 flashcard audio clips at 40 characters: MAI Flash vs Sume

Three thousand flashcard clips of 40 characters each are 120,000 characters of speech. At list that costs $1.80 on MAI-Voice-2.1-Flash, $2.64 on MAI-Voice-2.1 and $5.70 on Sume TTS. A language-learning app can voice a whole deck for the price of a sandwich.

The bill

The number that matters for a flashcard deck is the file count. Each card needs its own tiny audio file, so the job design decides the effort.

Jobs of 300 cards

A Sume job takes up to 20,000 characters, so 3,000 cards would need six jobs on the character cap alone. The audio limit is the tighter one, as the section after next shows. Put one card per sentence, ask for sentence segmentation with emit_audio: true and a wav container, and each card returns as its own hosted file. Without the wav or raw container you get only timings, so set it from the first request.

Keep every card as a full sentence that ends in a period. A single word with no punctuation may merge with the next line, and a merged slice means a card with two words on it. Number the cards in your own table and store the job id and zero-based segment index next to each, so a later edit finds the right file.

Mind the 1,200-second limit

A 40-character card is about three seconds of speech, an estimate you should check. At that length 500 cards run 1,500 seconds, which is over the 1,200-second limit and would fail with tts_duration_exceeded. Batches of 300 cards, about 900 seconds and 12,000 characters, fit, so the deck is 10 jobs.

On Flash a call is one clip, so 3,000 clips are 3,000 calls. The calls are cheap, but 3,000 separate responses each need naming and storing, while six Sume job results already contain the hosted files.

What the table leaves out

Microsoft's figures are the launch prices from 2026-10-01 (Microsoft AI, read 2026-10-05): $22 per 1M characters for MAI-Voice-2.1 and $15 per 1M characters for MAI-Voice-2.1-Flash, which makes up to 45 seconds of audio per call. Sume's figure is its list rate per transcript character, and on Sume spaces and punctuation count toward usage, so count characters of the final text.

The cheaper per-character price does not make the Microsoft route a drop-in swap. A Sume TTS request is an async job whose result carries a hosted audio_url, duration, optional word timings and sentence slices, and that hosted file is what Timeline audio and renders accept. Count the extra work of storing and joining files outside Sume before you call the gap a saving.

Check language coverage first. Microsoft lists 23 languages and 26 locales for MAI-Voice-2.1, and Sume takes a language field that you must set for every non-English transcript. A cheaper model that does not speak your target language is not an option, so settle that before you compare totals.

Edits and languages

Decks change. When a card text is edited, re-run only the job that holds it, not the deck. At 300 cards a job, one re-run costs about $0.57 on Sume at list. If you expect many edits, use smaller jobs of 100 cards at roughly $0.19 each and keep the index table by job.

Language matters for a language deck. Set language on every request, because the field decides which voices fit, and expect a warning when the voice and language do not match. Pick a voice per language and keep it, so each deck sounds the same each time you rebuild it.

Finally, dry-run the first batch with the quote option before you spend anything. The quote shows the exact character count of the 300 cards you send, and a cap on spend per job stops a malformed batch, for example a deck file with stray markup, from running up a larger bill than you planned.