MAI-Voice-2.1 at $22 per million characters vs Eleven v4

On the prices listed today, MAI-Voice-2.1 and Eleven v4 cost the same: $22 per million characters, or $0.022 per 1K. The Eleven figure is a promo (72% off through Oct 12), so the tie may end in a week. MAI-Voice-2.1-Flash is cheaper at $15 per million, and Eleven v4 Turbo is $11 per million in the same promo. They differ more in reach: MAI lists 23 languages and 26 locales, Eleven v4 states 90+ languages.

Price per million characters

Conversion: Eleven lists per 1K characters, so $0.022 x 1,000 = $22 per 1M and $0.011 x 1,000 = $11 per 1M. Sume's Sonic line is the Cartesia list of $38 per 1M credits (one credit per character) x 1.25 = $47.50 per 1M, before the 5.5% platform fee.

What else the Microsoft post says

• MAI-Voice-2.1 offers one voice across all its languages with native accent, and cloning from just a few seconds of reference audio with consent guardrails.
• MAI-Voice-2.1-Flash is described as 55% faster inference and about 60% cheaper. That cannot be the gap between the two list prices above, which is 32% ($15 vs $22), so the baseline must be something else; the text I read does not say.
• Availability listed: OpenRouter, Microsoft Foundry, MAI Playground and Vercel, with LiveKit coming soon.

What this means if you build on Sume

Sume does not ship MAI-Voice or Eleven. Its TTS Router serves Cartesia Sonic models, billed per character with each job rounded up to the cent. A 20,000-character job, the per-request maximum, is $0.95 on Sume against $0.44 on MAI-Voice-2.1 at list ($22 x 0.02) and $0.30 on Flash ($15 x 0.02). The gap is the price of the managed route: Sume hosts the artifacts and tracks each job. If raw per-character cost is your only constraint, go to the vendor directly.