Which Sume route transcribes a 5-minute video? Three compared
Video inspect, detach plus STT, and legacy video analyses all return speech. For 5 minutes: $0.05 plus compute, $0.06, and $0.30 on production only.

For a 5-minute video, use video inspect with transcribe: true when you also want stills, or audio detach plus STT when you want only the words: about $0.05 plus compute versus $0.06 flat. Legacy video analyses also returns speech, but it costs $0.30 and is retired on dest with a 410.
The three routes
Inspect captures its own container seconds times the Modal list rate times 1.25, plus a platform fee, so its total is not a flat number. The page does not quote a compute figure, so price it from one real run rather than from a guess.
| Route | Returns | Cost for 5 minutes | Status |
|---|---|---|---|
POST /v1/video-inspect with transcribe: true | Probe, stills, transcript with words and optional sentence segments | $0.05 for STT plus the inspect's Modal compute | Default on dest and prod |
POST /v1/audio-detach then POST /v1/stt-1.0/transcribe | Audio file, then words and optional sentences | $0.01 + $0.05 = $0.06 | Current |
POST /v1/video-analyses with include_transcript: true | Typed scenes with per-scene speech | $0.30 | Prod until PR-C2; dest returns 410 video_analysis_retired |
Pick by what you need next
- You need words for captions or a script: detach plus STT. The cost is exact, and you get a reusable wav.
- You need a quick look and the words together: inspect. It returns 8 mid-bin stills by default, up to 24 per call, plus the transcript.
- You need scene-by-scene structure: legacy analyses is the only route that returns typed
scenes[], and the page says not to start new work on it. On dest, usevideo_analyzeorvideo_segmentonly if they appear intools_list.
Limits to know before you queue
Inspect and detach read only media.sume.com clips that your workspace owns, so import first with POST /v1/media-imports. Both cap the source at 1,800 seconds. Detach outputs at most 900 seconds, and the STT duration hint tops out at 10 minutes.
Analyses has a soft cap of about 90 seconds, past which a job still succeeds but can carry the warning low_confidence_long_video, and a hard cap of 300 seconds. A 5-minute video sits exactly at the hard cap, and the page also warns that longer videos give less accurate results. That is a second reason to avoid it for a 5-minute file.
A silent clip fails inspect with inspect_source_has_no_audio and detach with detach_source_has_no_audio. Probe first with frames: false, which is enough to read probe.has_audio.
Why the legacy route still shows up
Older guides and saved scripts still point at video-analyses. If a script of yours gets a 410 on api.dev.sume.com, that is the dest flag, not an outage. On production the route stays on until its retirement step, but a new integration should not depend on it. Move to inspect and detach, which behave the same on both environments.
A worked choice
Suppose a marketing team has twelve 5-minute webinar clips and wants subtitles and a text summary. Detach plus STT is $0.06 a clip, $0.72 in all, and the transcripts feed both jobs. Inspect would add stills they do not need, and legacy analyses would be $3.60 for the batch on production and a 410 on dest.
If the same team wanted a contact sheet of eight frames per clip as well, inspect becomes the better single call, because the stills and the transcript come back together and there is no second job to wait for.
Checking the ceiling on yours
Before running a batch, read one result and note the fields you actually use. If your code only reads words, the other two routes add nothing. If it reads scenes[], you are on the legacy path and should plan the move now, because the dest environment has already switched it off and production will follow.
Sources
Related posts
More in Developers
- Which Sume spend guard applies: Completions, Formats, schedules, MCP
One table of the spend and retry guards on each Sume surface, for teams that put a new LLM or decision model in front of paid calls.
- Which Sume video models take 30 seconds? Filter /v1/videos/models
Seedance 2.5 and Wan 3.0 list 30 s in supported_durations. A 12-line Python script reads GET /v1/videos/models and prints every id that accepts your length.
- Which Sume video rows take video_url: Omni edit, Recast or Genjutsu
Three Sume rows take a source video_url: Gemini Omni edit ($0.125 a second at 720p), H3 Max Recast ($0.375 and up) and Genjutsu ($0.3975 and up). Pick by task.
- Which URL goes into video-captions after a trim: the new artf_
After video-trim, caption the new video_url from the trim result, not the source render. It is a new artifact on media.sume.com and a public HTTPS URL.
Written by Sume