Answer: MiniMax H3 at 480P is the cheapest video in this catalogue at $0.01 per output second (1 credit). Seedance 2.0 Mini is next at $0.02 (2 credits) and is the cheapest route that also accepts real-face references.
Both figures are per output second, not per clip, and both come from the same credit pool on Callirra — one credit is $0.01. That matters because "cheapest per second" is the only price comparison that survives a change of duration.
The per-second table
| Model · tier | Credits / output second | USD / second | 10-second clip |
|---|---|---|---|
| MiniMax H3 · 480P | 1 | $0.01 | $0.10 |
| MiniMax H3 · 768P | 2 | $0.02 | $0.20 |
| Seedance 2.0 Mini · 480P | 2 | $0.02 | $0.20 |
| Seedance 1.0 Lite · 480P | 2 | $0.02 | $0.20 |
| MiniMax H3 · 1080P | 3 | $0.03 | $0.30 |
| Seedance 1.0 Lite · 720P | 5 | $0.05 | $0.50 |
| Seedance 2.0 Fast · 480P | 6 | $0.06 | $0.60 |
| Seedance 2.0 · 480P | 10 | $0.10 | $1.00 |
| MiniMax H3 · 2K | 10 | $0.10 | $1.00 |
| Seedance 2.5 · 480P | 14 | $0.14 | $1.40 |
Higher tiers climb fast: Seedance 2.0 at 4K is 105 credits per second ($1.05), Seedance 2.5 at 720P is 32 ($0.32). Those are not mistakes — they buy resolution and instruction-following, not a cheaper second.
Why "per second" is the right unit
A clip is not a fixed product. The same model at the same tier can render 4 seconds or 15, and most platforms price by duration. Comparing "cost per clip" tells you nothing when one tool's default is 5 seconds and another's is 10.
Comparing per output second normalises duration, so the only remaining variables are resolution and model family — which is exactly what you want to compare.
Where the "cheapest" claim breaks
Cheapest per second is not cheapest per usable second. If a five-second draft at $0.01/s is unusable, its real cost is infinite until you pull again. Budget as: pulls × seconds × rate ÷ seconds you keep. A model that costs 3× more but lands in one take is cheaper.
The cheap tier caps out. H3 480P runs to 15 seconds; standard 1080P stops at 10. Seedance 2.0 Mini stops at 720P. If the deliverable needs 1080p, the "$0.01" tier is not an option at all.
Free tiers change the arithmetic. Both models are usable on a new account's free credits: 18 in week one, 15 a week after — enough for 18 seconds at H3 480P or 9 seconds of Seedance Mini 480P.
What I would actually buy
- Drafts and blocking: MiniMax H3 480P at $0.01/s. Twelve five-second explorations cost 60 credits.
- Real faces on a budget: Seedance 2.0 Mini at $0.02/s, because the standard Seedance route documents real-face references.
- A deliverable: H3 1080P at $0.03/s, which is cheaper than Seedance 2.0 Fast 480P.
- 4K video: none of these. Check 2K/4K lines, which are 10–105 credits per second.
FAQ
Is there anything cheaper than $0.01 per second? Not in this catalogue. One credit per output second is the floor.
Does the price include audio? On H3, yes — every output carries a native audio track, and reference audio is free. On Seedance, the audio switch is part of the request, not an add-on.
Do these prices change? They can. The numbers above are read from the live credit table, and the site shows the exact quote before a job runs.
Is the cheapest tier good enough to publish? At 480P, usually not for delivery; it is good enough to decide. Finish the selects at 768P or 1080P.
The live table, with every model and tier, is on Callirra's credit-cost page. If you find a cheaper per-second lane anywhere — including a reseller — the useful comparison is the same resolution and duration, not the headline rate.
Top comments (4)
"Per output second" is the right normalization, and the table is internally consistent - but two properties of the unit itself still hide outside it.
The first is billing granularity. The table assumes continuous per-second billing: a 4-second clip costs 4x the rate. If the platform rounds per call (bill each started second, or a 1-second minimum per generation), short exploratory clips pay a rounding tax that the per-second column does not show, and the tax is largest exactly on the cheap tier where drafts live. One line in the table's header - "billed per started second" or "exact duration" - would settle which world the reader is in.
The second is what happens to credits on a failed or moderated generation. Your budget formula (pulls x seconds x rate / seconds kept) is the right frame, and the pulls term is where the real money goes. But it behaves differently depending on whether a rejected pull refunds: if failures consume credits, the formula holds as written; if they refund, the effective cost drops for everyone, but more for the models with higher rejection rates. Either policy is fine, and they produce different rankings - worth one column.
Neither criticism touches the numbers you printed. It is the same lesson as your closing warning about comparing resolvers at the same resolution and duration: a price table can be perfectly accurate and still not be a ranking, because the effective sort key is rate times expected pulls, and pulls is the one variable the table cannot contain.
The 'cheapest per second is not cheapest per usable second' section is the important one; pulls x seconds x rate x seconds you keep beats any rate card.
Noting that Seedance 2.0 Mini is the cheapest route that also takes real-face references is the kind of constraint that decides the model before price does.
The per-second unit is the right call — cost per clip tells you nothing when defaults range from 5 to 15 seconds. The H3 vs Seedance gap is obvious in the table.