Sume Format run spend cap: 0 is rejected, null means $500

On a Sume Format run, generation_spend_cap_usd accepts a number up to 500; null means the platform maximum of $500; 0 or a value above 500 returns 400. If you omit it, the run inherits the Format's own cap, and a Format that never set one uses the platform default of $400.

The rules are in the Calling a Format docs. The reasoning given there is that a run that cannot spend cannot deliver, so a zero cap is refused instead of silently failing later.

What each value does

The effective cap shows on every receipt as usage.generation_spend_cap_usd_micros, in millionths of a dollar, so 120 dollars reads as 120000000.

A cap above the Format's own cap

Sume accepts a number higher than the Format's cap and does not clamp it. A Format saved with a $50 cap can therefore run with generation_spend_cap_usd: 200 if the caller asks. That makes the field a spending decision for the caller, so set it from your own budget and not from a copied sample. null lifts the ceiling to $500 but does not remove it.

{
  "instruction": "Make a 20 second promo from the attached product photo",
  "generation_spend_cap_usd": 25
}

Sizing a cap

Pick a cap from what one run can honestly cost. If a Format makes one 10-second clip at a price from the public catalog, multiply by the number of takes you allow, then add a margin. For example, 3 takes of a 10-second 720p wan-3.0 clip at $0.125 per second is 3 x 10 x $0.125 = $3.75, so a cap of $5 leaves room for a retry and stops a loop long before $400.

• Bulk items can bind their own cap, so a 20-row batch can have 20 different ceilings.
• A run that reaches its cap stops spending; read the receipt for the outcome.
• The cap limits generation spend, not the other fees of a run.

Putting a cap on a batch

A bulk queue sends each item as a normal Format run, and each item can bind its own generation_spend_cap_usd. Use that to size the cap per row: a row that needs one image can have a $1 cap while a row that makes a 30-second video has $15. A queue of 20 rows with caps of $5 each can spend at most 20 x $5 = $100 on generation, which is the figure to compare with your wallet before you submit.

Remember that workspace generation concurrency and wallet checks still apply to every child, so a low balance shows up as failed items and not as a failed create.