An agent makes 7 vertical 6-second clips on Seedance 2.5: $11 to $60
Seven 6-second 9:16 Seedance 2.5 clips cost $11.34 at 480p, $24.29 at 720p and $59.71 at 1080p on Sume. Run dry_run first and set max_spend_usd on each call.

If you ask an agent to make seven 6-second 9:16 clips on Seedance 2.5 through Sume's hosted MCP, the generation spend is $11.34 at 480p, $24.29 at 720p and $59.71 at 1080p. Each clip is billed at the catalog price times 1.25, rounded up to the cent. Set max_spend_usd just above the total you expect, and run dry_run before the first paid call.
Price per clip and for seven
The per-clip figures come from Sume's video catalog data, which prices Seedance 2.5 by video tokens at list times 1.25 and rounds up to a cent. The Video Router and Video generation pages describe the 4 to 30 second range for seedance-2.5 at 480p, 720p and 1080p, and the 9:16 aspect ratio.
| Resolution | Per clip | Seven clips | Arithmetic |
|---|---|---|---|
| 480p | 162 cents ($1.62) | $11.34 | 7 x 162 = 1,134 cents |
| 720p | 347 cents ($3.47) | $24.29 | 7 x 347 = 2,429 cents |
| 1080p | 853 cents ($8.53) | $59.71 | 7 x 853 = 5,971 cents |
What to tell the agent
Name the model, because leaving it out routes to sume/auto, which is a different price. Give each clip its own idempotency_key, such as launch-clip-01 through launch-clip-07, so a retry after a dropped connection cannot create an eighth job. Then wait for the whole wave with one batch jobs_wait of seven ids instead of seven single waits.
- Get the model ids from
video-router_models, not from memory. - Run
dry_run=trueon the first clip and compare the estimate with the table. - Set
max_spend_usdper call; Sume enforces it only when you provide it. - Repeat
jobs_waitonwait_slice_expired; the cap per call is 55 seconds.
The plan matters at seven
A Free workspace accepts 6 paid jobs at once (1 processing plus 5 queued), so the seventh submit would return 429 queue_full. Pro accepts 24, which covers seven clips with room to spare. If you are on Free, submit six, wait for one to finish, then submit the seventh with its own key. Only one job runs at a time on Free, so the wave takes longer. Pro processes four at a time.
| Plan | Processing | Queued | Accepted |
|---|---|---|---|
| Free | 1 | 5 | 6 |
| Pro | 4 | 20 | 24 |
| Startup | 8 | 40 | 48 |
Cheaper ways to the same storyboard
Draft at 480p first. The seven drafts cost $11.34, and you pay the 720p price only for the clips you keep: for example, 3 keepers at 720p add 3 x $3.47 = $10.41. If the length can be shorter, remember that billing scales with seconds, and the 4 second minimum for this model is $2.32 at 720p (232 cents).
Report the real spend
When the wave finishes, have the agent call usage_get and quote the debited amount per job, not the estimate from dry_run. The two should match for a completed clip. A failed job releases or refunds its reservation, so the final bill can be lower than the table if some clips fail. Ask the agent to list the job ids, the status of each, and the total debited, and to give the media.sume.com URLs rather than signed links.
If the total looks wrong, read jobs_events for the job in question before you resubmit. Resubmitting a paid clip because a local wait timed out is the one mistake that can double the bill.
Sources
Related posts
More in Agents
- Agent turns a product page into a clip: crawl_scrape, generate_video
On Sume MCP, crawl_scrape is a read tool that works with Write off. generate_video is paid and needs Write or a key. Split the job in two sessions to cap spend.
- 12 images from an agent on GPT Image 2.5: 12 cents at low, 72 at high
On Sume, GPT Image 2.5 bills 1 cent at low, 2 at medium, 6 to 7 at high per 1024-class image. Quality defaults to high: 12 images cost 72 to 84 cents.
- Backend job that needs an agent: Agent Completions or hosted MCP?
Agent Completions runs Sume's agent and returns a 202 receipt with a required spend cap. Hosted MCP is for a client whose own model calls Sume tools.
- Swap the LLM behind your MCP client: do Sume's gates change?
No. Hosted MCP gates sit in scopes, idempotency_key and max_spend_usd, not in the client model. What each session type can see, whatever model it runs.
Written by Sume