AI video analysis API cost: free probe and stills, $0.01 a minute

Analyzing one clip with Sume's video inspect costs nothing for the probe and the stills, and $0.01 per audio minute if you ask for a transcript. The docs cap the duration hint at 600 seconds, so one transcript call reserves at most $0.10.

It is not a semantic analysis. Inspect returns facts about a clip, the frames you ask for and optionally words; it does not answer questions about the content. This post covers what each part costs and how to avoid paying for a transcript you cannot use.

What does video inspect bill, and what is free?

Probe and stills are free, so the batch cost is entirely the transcript line: audio minutes times $0.01. The numbers below are arithmetic on the published rate. For the 30-second rows, no duration_seconds hint is sent, so each call reserves the documented one-minute default; for the longer rows the hint is set to the real length.

That default is why 100 short clips come to $1.00 and not $0.50. If most of your clips are short and you omit the hint, the one-minute reservation is the real price.

What does it cost for a batch of clips?

Probe and stills are free, so the batch cost is entirely the transcript line. Multiply the audio minutes you transcribe by $0.01. The numbers below are arithmetic on the published rate, rounded up per clip, with the hint set to the real duration.

Each 30-second clip rounds up to one minute, which is why 100 clips cost $1.00 and not $0.50. If most of your clips are short, the per-call minimum is the real price.

How do I avoid paying to transcribe a silent clip?

A silent clip fails with inspect_source_has_no_audio, so the first call should be a free one. The docs recommend checking probe.has_audio with a frames: false inspect before asking for a transcript.

The request below does that check. It is a probe only, so it should not bill.

curl -X POST https://api.sume.com/v1/video-inspect \
  -H "Authorization: Bearer $SUME_API_KEY" \
  -H "Content-Type: application/json" \
  -H "Idempotency-Key: probe-001" \
  -d '{"video_url":"https://media.sume.com/artifacts/artf_demo/clip.mp4","frames":false}'

What about clips longer than ten minutes?

The source limit is 1,800 seconds but the transcript duration hint tops out at 600. The docs do not say how a longer transcript is billed, so I will not guess. If you need a long clip transcribed with a known cost, cut it into ranges of 10 minutes or less with video trim at $0.02 per job and inspect each part.

For a 30-minute recording that is three trims ($0.06) plus three transcripts ($0.30), so $0.36 in total at published rates.

When is inspect the wrong tool?

If you need exact frames at a time, use video frames, which is also unbilled. If you need typed scenes or semantic answers, inspect does not provide them. The docs point to other surfaces for that and say availability differs by environment.

Inspect fits the cheap first look: check duration, audio and a few stills before you commit to a paid generation or a longer pipeline.

How should I budget a pipeline that inspects first?

Treat inspect as a gate. A free probe tells you the duration, whether there is an audio track and how big the source is. Only clips that pass your rules need a transcript, and only transcripts you will read need to be paid for.

A reasonable order is: probe with frames: false, sample a handful of stills if you need a visual check, then transcribe the clips that have audio. On a library where a third of the clips are silent screen recordings, the gate saves a third of the transcript line, and it saves the failed call as well.

Keep the job id from each call. Inspect jobs appear in the usage ledger like any other, and a scoped read by job id shows what one call cost.