MiniMax video tasks are queryable for 7 days: poll every 10 s
MiniMax says video tasks can be queried for the last 7 days and suggests polling every 10 seconds. What that means for your pipeline, and Sume's 30 s example.

MiniMax's H3 guide says a video task can be queried for the last 7 days and recommends polling every 10 seconds (read 2026-10-02). So copy the finished video to your own storage within a day or two, not at the end of the week. Sume's /v1/videos docs show a 30 second polling interval in their examples and return an unsigned_urls entry per finished job.
What does the MiniMax flow look like?
The guide describes three steps: create a task and receive a task_id, poll its status, then retrieve the video from the returned URL. The guide lists final states succeeded, failed and cancelled. The query reference for the older Hailuo endpoint lists a different set, Preparing, Queueing, Processing, Success and Fail, and returns a file_id on success (both read 2026-10-02). Write your status handling per endpoint, not per vendor.
| Item | What the page says |
|---|---|
| Recommended poll interval (H3) | 10 seconds |
| Task lookup window (H3) | Last 7 days |
| Final states (H3 guide) | succeeded, failed, cancelled |
| Final states (Hailuo query reference) | Success, Fail |
| Who can query a task | Only tasks created under the current account |
What does Sume do?
On Sume you submit to POST /v1/videos, receive a job id and polling_url, and poll until completed, failed or cancelled; pending and in-progress are the middle states. Per the video generation docs, the example loop sleeps 30 seconds between polls, and the best-practices section says video typically takes 30 seconds to several minutes depending on model and parameters.
The same job is also readable at GET /v1/jobs/{id}/status and /result, and a callback_url removes the need to poll. Sume's pages used here do not state how long a finished job stays queryable, so do not treat the 7 days as a Sume guarantee; download the artifact when you first see completed.
How should you choose an interval?
- Direct to MiniMax H3: 10 seconds is the vendor's own number; do not go faster.
- On Sume: 30 seconds suits clips that take minutes; poll less often for long renders.
- With many open jobs, polling cost scales with job count, so prefer webhooks above a few dozen.
- After a restart, re-query by stored job id; never resubmit a paid create to find out.
What to store
Keep the job or task id, the prompt, the model id, and the downloaded file. With MiniMax, the id plus the 7 day window is all you have to recover a missed result. With Sume, send an Idempotency-Key on create so a retry returns the original job instead of billing a second one.
Sources
Related posts
More in Models
- Where are the lyrics in a Lyria 3.5 result? Gemini vs Sume
Google returns Lyria 3.5 lyrics and song structure as text beside the audio. Sume puts model-reported lyrics or a section map in result.lyrics when present.
- Nano Banana 2 Lite model id: Vertex hyphens vs Gemini API dots
Google's Gemini API names Nano Banana 2 Lite gemini-3.1-flash-lite-image; its Cloud post writes gemini-3-1-flash-lite-image. Sume's image catalog lists neither.
- Nano Banana negative prompt: describe what you want instead
Google's Gemini image guide says to write semantic negative prompts: describe an empty street, not 'no cars'. Sume's image request has no negative field.
- Nano Banana prompt languages: Korean and Japanese, per Google
Google lists the languages Gemini image models work best in, including ko-KR and ja-JP. What that means for a Korean or Japanese prompt sent through Sume.
Written by Sume