Grok Voice Transcribe 2.0 price per hour vs Sume STT per minute

xAI's models page lists speech to text at $0.10 per hour in batch mode, and its release notes say grok-voice-transcribe-2.0 became available on September 17. Sume STT is billed per audio minute at $0.01 per audio minute; if you omit duration_seconds it reserves 1 minute, and the documented maximum is 10 minutes.

xAI figures are from its models page and release notes; Sume's from the API contract and Usage, read 2026-10-01.

What does xAI list for transcription?

The models page shows "STT (Batch) $0.10 / hour" and "STT (Streaming) $0.20 / hour". The release note says to use grok-voice-transcribe-1.0 or grok-voice-transcribe-2.0, and that the default is grok-voice-transcribe-1.0. Dividing the batch rate by 60 gives about $0.0017 per minute, which is my arithmetic, not a number on the page.

How does Sume bill STT?

Per audio minute, with the estimate reserved first. The usage ledger defines reserved as "Estimated usage was reserved before provider execution", then captured after success and refunded if the reservation is released after failure or cancellation.

What does the 10 minute cap mean for long audio?

The duration_seconds field, which the API describes as the audio duration used for usage reservation, has a maximum of 600 seconds (10 minutes). For longer recordings, split the audio into pieces of 10 minutes or less to stay inside that documented maximum. Send duration_seconds when you know it so the reservation matches the audio instead of defaulting to 1 minute.

Can I pick a Grok transcription model on Sume?

The docs I read do not list a Grok transcription model for Sume STT, so do not assume model ids carry over. For batch patterns on Sume, see batch transcription API.