Sume's new API announcement raises a practical question: what will a Seedance 2.5 shot actually cost? Its public model page currently lists $0.1028 per output second at 480p, $0.2312 at 720p, and $0.5686 at 1080p. The useful comparison starts with the exact request and ends with the footage you can keep.
This explanation uses Sume's public pricing and model interfaces reviewed on October 23, 2026. Seeora has not run a paid comparison of Sume against other providers, measured its queue times, or integrated its API. The calculations below illustrate budgeting from published rates; they are not generation results.
AI-generated concept illustration made with Codex imagegen, not an application screenshot or a Seedance output.
What Sume announced
On October 22, David Im announced Sume API, describing a service with no routing markup and comparing its Seedance pricing with other providers. A follow-up described pay-as-you-go access without subscriptions. These are provider statements. They do not independently establish the lowest available price or equivalent performance across services.
The new API pricing interface displays its own rates beside competitor figures. Its footnote dates the competitor snapshot to October 16 and says the Higgsfield column uses list prices before launch discounts. That qualification matters: a comparison against a dated list rate does not describe every customer's current bill.
“Zero markup” also describes Sume's commercial claim, rather than free generation. You still pay the published rate for a completed request.
Seedance 2.5 rates and a ten-second example
The Seedance 2.5 model playground lists these rates and says requests round up to $0.0001. Its default five-second, 480p example displays an exact charge of $0.5140 before running.
| Output option | Published rate per second | Ten-second arithmetic |
|---|---|---|
| 480p | $0.1028 | $1.0280 |
| 720p | $0.2312 | $2.3120 |
| 1080p | $0.5686 | $5.6860 |
The last column multiplies the displayed rate by ten. It does not include another model, an editing service, or a separate finishing step. Confirm the estimate shown for your actual inputs before submitting, especially when changing reference mode or other options.
Save the model identifier, duration, resolution, aspect ratio, audio setting, input mode, displayed estimate, and observation date together. A price copied without its configuration is hard to compare later. The playground describes text, image, and reference generation, 480p–1080p output, four-to-thirty-second clips, and native audio; verify the controls exposed by the route you intend to use.
Read the current API surface carefully
Sume also has an older comparison guide dated September 24 that describes routed video pricing with a multiplier and an agent fee. The newer API interface reviewed for this article presents the no-markup offer and an exact per-request estimate. These pages describe different terms and dates; copying an old formula into a budget for the new interface would create an unexplained mismatch.
Start from the current model page, follow its documentation, and check the account's applicable billing terms. If the estimate and written terms disagree, resolve that difference before a batch run. This article does not determine which historic terms apply to an existing account.
Refunded failures and rejected takes are different
Sume's API pricing FAQ says it holds the price when a request starts, captures it when the request finishes, and refunds failed or canceled requests in full. A completed video that you dislike is a different outcome. The FAQ does not promise a creative-quality refund for a shot with the wrong expression, an awkward hand, or an unusable camera move.
Track technical failures separately from completed takes rejected during editing. For example, suppose three ten-second 1080p requests complete and only one fits the scene. At the table's published rate, those requests total $17.0580 for one usable shot. This is hypothetical arithmetic, not an observed acceptance rate.
Record why each completed take was rejected. Identity drift calls for different changes from incorrect timing. Repeating the same request without recording the failure reason can make a lower per-second rate surprisingly expensive.
Budget the experiment before the final render
Choose one short scene and define what would make it usable: the subject keeps the intended appearance, the important action finishes, the camera reveals the right object, and the required sound is present. Then set a maximum number of completed attempts and a spending ceiling.
A planning example is one ten-second 480p attempt followed by one ten-second 1080p attempt. The displayed rates imply $6.7140 if both complete. Two completed 1080p attempts imply $11.3720. This shows the arithmetic of allocating an exploratory attempt; it does not prove that a low-resolution result will transfer unchanged to the higher-resolution generation.
Keep the prompt and references organized so the second request has a clear purpose. If the first result reveals a structural problem, fix that problem before increasing resolution. For the related preview-versus-finishing distinction, read the Seedance 2.5 Draft Mode explanation; a provider's ordinary resolution selector should not automatically be treated as that workflow.
Check concurrency before automating a batch
The API pricing FAQ says concurrency depends on funds added during the previous 28 days. It lists tiers of 2, 10, 20, and 40 at the displayed funding thresholds, and separately says a first API key starts at 4. Check your account's effective limit rather than assuming a universal default.
Images and videos share the limit, and queued and running requests both count. The FAQ says requests above the limit are refused without a charge. An application should therefore control outstanding jobs and inspect their status before launching more work. A rate card cannot tell you the completion time or reliability you will experience.
What to record before switching providers
For one representative scene, keep a small ledger: configuration, estimate, final charge, completion status, elapsed time, and whether the clip passed your review. Compare the same required output on both services, including audio and reference needs, rather than comparing unrelated cheapest options.
The next useful step is a bounded trial with your actual scene and acceptance criteria. Review the delivered file and the settled charge together. Sume's new rates make a concrete option to evaluate; your cost per usable shot depends on the settings, completed retries, and finishing work your project requires.