Batching saves money when many small jobs would each pay a submission minimum. It does not change the list fare for work that already clears that minimum on its own.
One submission, one minimum.
Provisional fares. These are the published limited beta fares. Check the fare table before submitting work.
Each submission with at least one successful item pays the fare for its billed units or the $0.0075 minimum per submission, whichever is larger. A prediction is a batch of one, so it has the same minimum. Failed and cancelled items cost nothing; if nothing succeeds, the whole submission costs nothing.
That is why a small catalog suits the queue. An image under 1 MP still bills one unit. Multiple successful images share the submission minimum when you submit them together.
Compare the same image work.
| 100 images, each at 1 MP or under | One batch, list fare | 100 separate submissions |
|---|---|---|
| Real-ESRGAN x4plus | $0.10$0.0075 minimum per submission. | $0.75$0.0075 minimum per submission. |
| BiRefNet general | $0.05$0.0075 minimum per submission. | $0.75$0.0075 minimum per submission. |
At list fare, the minimum clears at 8 Real-ESRGAN units or 15 BiRefNet units. Larger images contribute more units: for example, a 2 MP input contributes two units.
A lone job here costs more than on Replicate. Our lone image costs $0.0075, with a $0.0075 minimum per submission. Replicate lists $0.002 per image; Runware lists about $0.0013 at 4x, varies with input and also costs less for a lone image. Competitor prices checked 2 Nov 2026.
A lone job here costs more than on Replicate. Our lone image costs $0.0075, with a $0.0075 minimum per submission. Replicate lists about $0.0017 per run; Runware lists $0.0006 per image and also costs less for a lone image. Competitor prices checked 2 Nov 2026.
Audio has a separate floor for every file.
A file under 30 seconds bills as 30 seconds. This 30-second minimum per file applies before units are added across the batch. Twenty one-second clips bill 10 audio minutes, not twenty seconds.
| Whisper work | One batch, list fare | Separate submissions |
|---|---|---|
| 20 one-second files | $0.0075$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. | $0.15$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. |
| 20 five-minute files | $0.044$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. | $0.15$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. |
| 20 one-hour files | $0.528$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. | $0.528$0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. |
The one-hour files pay the same list fare either way because each file already clears the submission minimum. The one-second files still pay their per-file floor in a batch. A finished transcription with an empty transcript remains billable.
DeepInfra beats us on a short Whisper job. Our lone one-minute file costs $0.0075, with a $0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. DeepInfra lists $0.00045 per minute. Competitor prices checked 2 Nov 2026.
Check launch eligibility at settlement.
A launch seat and enough successful billed units are both required for the launch fare. Declaring a large max_units does not qualify a small or failed batch. The fare table explains the eligible pack, seat assignment, and remaining launch usage.
For 20 successful five-minute recordings, the list total is $0.044; an eligible launch total is $0.04. Both totals have a $0.0075 minimum per submission. A file under 30 seconds bills as 30 seconds. DeepInfra still costs less than our minimum for a short lone transcription.
Those 100 successful audio minutes pass Whisper’s 30-unit launch threshold. The image models currently have equal list and launch fares; a launch seat does not make those image jobs cheaper.
A hold is an upper bound, not the final bill.
The hold uses the submitted max_units at list fare. The final charge uses successful billed units and any eligible launch fare, never more than the hold. Unused hold returns to the account at settlement.
Size each item honestly, run dry_run, and inspect held_micros before accepting a batch. Poll until settled_at is set before treating charged_micros as final. See the sizing guide and dry-run example.
Expect a cold start of 10 to 30 s, occasionally minutes. This is a queue: check the batch state instead of assuming a fixed completion time.
Y-PAD is built for work that can wait in a queue. Published fares and competitor comparisons here use the catalog checked 2 Nov 2026; the linked comparisons show the providers who cost less for lone jobs.
Have a batch in mind?
Estimate its cost ↗

