12 MiniMax H3 clips at 768p and 13 s fit a $12 wallet: $11.70

A $12 balance buys 12 MiniMax H3 clips of 13 seconds at 768p on Sume. Each clip costs 13 x $0.075 = $0.975, so 12 clips cost $11.70 and leave $0.30, which is less than one more clip. At 480p the same wallet buys 14 clips; on H3 Max at 768p it buys 9.

Thirteen seconds is inside the documented 5-15 second window for both minimax-h3 and minimax-h3-max, so the length does not change the per-second price.

How many clips each option buys

The count is floor(12 / clip price). The leftover is what remains after those clips.

Why the leftover matters more than it looks

Sume reserves the estimated cost when it accepts a job. A balance of $0.30 cannot reserve another 768p clip of this length, so a thirteenth submit fails with 402 insufficient_credits rather than starting and running out partway. A failed or canceled job releases its hold, so the leftover can grow back if a clip fails.

Cancellation is limited. It succeeds only before generation work starts, so a job that is already processing will finish and bill. Do not count on canceling as a way to recover a hold on a long batch.

Queue limits for a 12-job batch

Twelve accepted jobs is under Pro capacity (24) and Startup capacity (48) but over Free capacity (6). On Free, send six, wait for some to finish, then send the rest. A seventh submit while six are accepted returns 429 queue_full, and Sume releases or refunds the reservation of that failed admission. Use an Idempotency-Key so a retry cannot create a duplicate paid job.

• 12 clips x 13 s x $0.075 = $11.70 at 768p
• 14 clips x 13 s x $0.0625 = $11.375 at 480p
• 9 clips x 13 s x $0.10 = $11.70 on H3 Max at 768p

Choosing between H3 and H3 Max at the same budget

The two H3 variants differ in price at 768p: $0.075 per second for minimax-h3 and $0.10 for minimax-h3-max. On a fixed $12 budget that is the difference between 12 and 9 clips of 13 seconds. H3 Max also offers 1080p at $0.20 per second, which on this budget is 4 clips ($10.40) with $1.60 left.

The Video API page describes H3 Max as the faster 768p variant with native stereo audio and image, video and audio reference inputs. If your clips need those inputs, the question is no longer price alone. If they are plain text-to-video drafts, the cheaper model buys three more clips for the same money. Neither price includes any extras beyond default options, so a request that adds paid options should be re-estimated before you commit the budget.

Checking the arithmetic yourself

Per-clip price is seconds x rate, and the count is the integer part of balance divided by that price. Keep the leftover in view, because it is the cheapest way to see whether a top-up is needed. A $0.30 leftover at 768p buys 4 seconds, which is below the 5-second minimum of both H3 models, so the money is stranded until the balance grows. At 480p the leftover after 14 clips is $0.625, which is enough for a 10-second clip at $0.0625 per second, so a fifteenth, shorter clip is possible if the prompt works at that length.