STT per hour: MAI 0.36, 0.10, 0.54 vs Sume 0.60 (read 2026-10-08)

On Microsoft's model page, batch MAI-Transcribe-2 is $0.10 per audio hour, MAI-Transcribe-2-Streaming is $0.54, and the older MAI-Transcribe-1.5 is $0.36. Sume STT 1.0 is $0.01 per audio minute, which is $0.60 per hour. So Sume costs 6.0 times the batch model, 1.11 times the streaming model and 1.67 times the 1.5 model. Both Microsoft 2-series prices are labelled introductory on the page, so the ladder can move; the Sume price is the catalog's current one.

The ladder in one table

The per-hour figures are as the Microsoft page states them. Per-minute figures are per-hour divided by 60, and the per-1,000-hour column is the per-hour price x 1,000. The language and streaming columns are the page's own fields; Sume's catalog entry states no language count, so I do not give one.

What the cheaper rows leave out

The streaming row is the price of a real-time feed, and the page's own feature rows show it with no word-level timestamps, no contextual biasing, no diarization and no transcription-style choice, while the batch row lists all of them. If you need word timings for captions, the batch model is both cheaper and richer than the streaming one on Microsoft's page. Sume's transcript carries word timings and optional sentence segments (Video inspect documents the shape), and its STT can take a language hint or auto-detect.

Sume STT is a job, not a stream: you submit audio, poll the job, then read the result, and one request is capped at 600 seconds (10 minutes). Microsoft's batch page figure is that one hour of audio finishes in about 10 seconds. Those are different products: pick Sume when the transcript feeds a Sume workflow such as a Timeline or caption pass and the hand-off matters, and pick the cheapest batch rate when you only need text.

Arithmetic for one Sume hour

A 60-minute recording on Sume is split into six 10-minute requests because of the 600-second cap, and 60 minutes x $0.01 = $0.60, the same as the table. Each request is well above the 1-cent job minimum in the catalog, so the floor does not add anything here. A 25-second clip is another matter: it is 0.4167 of a minute, but the catalog minimum of 1 cent applies, so tiny clips cost relatively more on Sume than the per-hour rate suggests.

A 1,000-hour sanity check

At 1,000 hours the Microsoft batch model is $100, the 1.5 model $360, the streaming model $540 and Sume $600. Sume's gap to batch is therefore $500 per 1,000 hours, or $0.50 per hour. Whether that is worth paying depends on whether you would have built the upload, chunking and polling around a different provider anyway; for a few hours a month the gap is cents and the integration cost dominates.