Groq Orpheus TTS: $22 per million characters and a 200-character cap

Groq's Orpheus V1 English model lists text-to-speech at $22.00 per million characters, and its model page says to keep input text under 200 characters per request (read 2026-10-06). It is English only. That is the same $22 per million that Microsoft lists for MAI-Voice-2.1 (digitalapplied tracker, read 2026-10-06), so the per-character price is a tie and the difference is how many requests you make.

Sume TTS 1.0 lists at $0.0475 per 1,000 characters, which is $47.50 per million, with a 20,000-character request limit. Sume is the more expensive of the two per character; the case for it is fewer calls, a job record and a receipt, not a lower rate.

Requests and cost by script length

What the cap means in practice

A 900-character narration needs at least 5 Groq requests, and the cuts have to land on sentence ends or the voice resets mid-thought, so you carry the splitting and the stitching. On Sume that script is one job. If audio is stitched from many requests, intonation can differ at each seam, which is a quality cost the table does not price.

Groq's page also describes bracketed vocal directions such as [cheerful]; check the page for the current set.

Pick by workload

• English only, short lines, lowest per-character price: Groq fits.
• Long narration in one take, a job id to poll, a webhook and a receipt, or a language other than English: Sume's single-job model is the simpler build.
• Do not read the table as a quality ranking. It is price and request arithmetic from list prices only.