How much money can Sume hold in flight per plan?
Accepted job capacity is 6 on Free up to 120 on Scale. At $2.50 per 10-second Wan 1080p job, a full Scale queue holds $300. Computed from the docs.

The most reserve a workspace can hold at once is accepted job capacity times the estimate per job. A full Free queue of 6 Wan 1080p 10-second jobs holds $15.00, and a full Scale queue of 120 holds $300.00. Success captures the reserve, failure releases it.
Accepted jobs by plan
Per the generation admission docs, each plan has a processing, queue and accepted limit. Accepted is the sum, and it is the number that matters for held money.
| Plan | Accepted jobs | Grok 10 s | Wan 3.0 1080p 10 s | Kling 3 audio on 15 s |
|---|---|---|---|---|
| Free | 6 | $0.75 | $15.00 | $18.90 |
| Pro | 24 | $3.00 | $60.00 | $75.60 |
| Startup | 48 | $6.00 | $120.00 | $151.20 |
| Scale | 120 | $15.00 | $300.00 | $378.00 |
| Enterprise | 120 | $15.00 | $300.00 | $378.00 |
How the hold works
At submit, Sume reserves the job's estimate. If the job succeeds, the reserve is captured as the charge. If it fails, or you cancel before it starts, the reserve is released or refunded. The ledger shows these as reserved, captured and refunded rows, and the usage summary exposes held_usd_micros for money still on hold.
If the balance cannot cover a new estimate, the request returns 402 insufficient_credits before any provider work begins.
- Held money is not spent money; it returns to the balance if the job fails.
- Cancel works only before generation starts.
- Read
held_usd_microsinstead of adding up rows yourself.
Sizing a balance
If you plan to keep a queue full, the balance should cover at least one full accepted queue of your most expensive job, plus whatever you expect to spend while the queue drains. For a Pro workspace running Kling 3 audio on at 15 seconds that is 24 times $3.15, or $75.60 held at once.
Reading the ledger
After a wave of jobs, the usage summary reports debited_usd_micros for captured money, held_usd_micros for reserves still open and refunded_usd_micros for released money. If held stays above zero long after the last job ended, look for a job that is still processing before you assume something is wrong.
Do not sum ledger rows by hand; the summary is already reconciled and states when the figure is final.
Limits
The tables show the model amount. A workspace Agent Fee line can be added on top, so treat the job's own usage cost as the truth and check the live catalog before you budget; limits and rows change. Plan limits can change; the generation_limits field is the authoritative value for your workspace.
Sources
Related posts
More in Pricing
- Kling motion control duration_seconds: it sets the credit hold
On Sume's Kling 3.0 Motion Control, duration_seconds (1 to 30) sizes the credit reservation; the output length follows the driving video, not this field.
- Minimum balance to submit a 10-second AI video job
A 10-second video job needs its estimate in your wallet at submit: $0.125 on Grok, $1.25 on Wan 720p, $3.75 on Omni Flash 4K, or a 402 comes back.
- Murf API at $0.01 a minute vs Sume TTS per character
Murf's API page lists $0.01 per minute. Sume bills text to speech per character. Here is the conversion to minutes, the arithmetic and what the gap means.
- Rask AI minute credits and 3x lip-sync vs a Sume dubbing pipeline
Rask bills 1 credit per video minute for Standard lip-sync, 3 for Enhanced. Sume has no dubbing endpoint; a chained pipeline runs about $0.55-$1.55 per 10 min.
Written by Sume