Gemini 3.5 Transcribe pricing per minute and the Sume STT rate

Google's pricing page lists Gemini 3.5 Transcribe at about $0.005 per minute, an estimate built from token prices. Sume's speech transcription is billed per audio minute at $0.01 per audio minute on the rate card, and reserves 1 minute if you send no duration_seconds.

This page compares only the billing unit and the reserve rule, not accuracy. Sources: Google's pricing page and Sume's Video inspect docs, read 2026-09-30.

Where does Google's $0.005 come from?

The page lists $2.00 per 1M input tokens (about $0.003 per minute of audio) and $12.00 per 1M output tokens (about $0.002 per minute of text). It says the estimate assumes 25 audio tokens per second for input and 175 text tokens per minute for output, for an effective blended rate of about $0.005 per minute. A different amount of spoken text changes the output share.

How does Sume bill the same minute?

Sume STT 1.0 is priced per audio minute after Sume margin; there is no separate output-token charge. Without a duration hint Sume reserves 1 minute, and the maximum is 10 minutes per request. The hint is capped at 600 seconds.

What does reserving 1 minute mean for me?

The reservation is an estimate held at submit; Sume's docs say successful completion captures the reserved usage. To avoid reserving less than the real length, pass duration_seconds for clips over a minute. For longer audio, split it; see transcribe long audio files.

Is there another way to get a transcript on Sume?

Yes: Video inspect with transcribe: true returns a transcript for a video, at a public rate of $0.01 per audio minute in the docs (confirm live in GET /v1/catalog). See transcribe video to text.