Bulk run spend cap: worst-case ceiling for 100 holiday items
generation_spend_cap_usd is per item, so 100 items at a $4 cap can spend up to $400. The platform maximum is $500 per item; 0 is rejected.

The spend cap on a Sume bulk run applies to each item, not to the whole queue. A 100-item queue with generation_spend_cap_usd set to 4 can therefore spend up to $400 in the worst case. The platform maximum per item is $500, a value of 0 is rejected, and null means $500.
Worst case by cap
Multiply the per-item cap by 100 for one queue and by 12 for a 1,200-item catalog. Docs as of 2026-10-08.
| Per-item cap | 100-item queue | 1,200 items (12 queues) |
|---|---|---|
| $1.00 | $100.00 | $1,200.00 |
| $4.00 | $400.00 | $4,800.00 |
| $10.00 | $1,000.00 | $12,000.00 |
| $50.00 | $5,000.00 | $60,000.00 |
| $500.00 | $50,000.00 | $600,000.00 |
Pick the cap from the model price
A cap that is near the model's real cost stops runaway items without cutting good ones. A 5-second Seedance 2.5 clip at 720p 9:16 is $2.889 (0.5778 x 5), so a cap of $4 leaves headroom for one retry-sized variation, while a cap of $2 would cut it. For 5 s on Wan 3.0 at 720p ($0.625), a $1 cap works.
- Cap below the price of one generation: items cannot finish.
- Cap at $500 (or null): no practical protection.
- The cap bounds spend, not queue time.
Other guards
The cap is one guard of three. The wallet reserve at submit and the plan's accepted capacity also limit exposure; see generation admission. Set the cap in the request for each item, review the first queue's real spend, and tighten the cap before you submit the rest.
Worked example
A catalog team plans 1,200 clips of Wan 3.0 at 720p, 5 seconds each, which is $0.625 per clip and $750 in total. With a $1 cap per item the worst case is $1,200 across the catalog, and the expected spend is $750. With a $0.70 cap, items still finish, since $0.625 is below it, and the worst case is $840.
Caps below the generation price mean items fail; caps above it give a ceiling. The right number is the one just above the real price.
- Expected: 1,200 x $0.625 = $750.
- Cap $1: ceiling $1,200.
- Cap $0.70: ceiling $840.
Before you scale
Review real spend after the first queue and tighten the cap before the next one. If the first 100 items average $0.63 each, the queue cost about $63, and a cap of $0.70 keeps every later queue within $70 unless an item runs long. Record the cap with each queue, so that an audit of a catalog run can show the ceiling that applied.
Sources
Related posts
More in Pricing
- Cartesia Startup and Scale vs Sume TTS: break-even in characters
Sume TTS costs $0.0475 per 1,000 characters. Cartesia's Startup plan breaks even near 1.03 million characters a month, Scale near 6.29 million. Math inside.
- Ceil to the cent on 3-second Wan clips: 37.5 cents becomes 38
A 3-second wan-3.0 clip at 720p is $0.375 after Sume's 1.25 factor, billed as 38 cents. Over 1,000 clips the round-up is worth $5.00.
- Cheapest AI video API per second: Sume models ranked by price
Cheapest per output second on Sume: Grok Imagine Video 1.5 at $0.0125, Gemini Omni 360p at $0.0375, then Wan 3.0 and MiniMax 480p at $0.0625. Ranked table.
- Cheapest Omni Flash test: 3 s at 360p is $0.1125 on Sume
Google lists no free tier for Gemini Omni Flash. The smallest legal request on Sume, 3 seconds at 360p with sound, costs $0.1125.
Written by Sume