Skip to main content
Strand AI bills predictions in credits. You can get the exact price of a run before you submit it.

​
How credits work

The model evaluates overlapping positions on an MPP-normalized grid. One credit pays for one retained tissue position on one marker:
1 credit = 1 retained tissue model-grid position × 1 marker.
The source-pixel footprint of each position depends on the sample’s calibrated microns-per-pixel (MPP). It is not a fixed 224×224 region of the uploaded image. Background positions are excluded, so a slide with one small biopsy costs far less than a slide of the same pixel dimensions covered in tissue. Raw byte size, raw pixel dimensions, and formulas such as ceil(width / 224) × ceil(height / 224) do not determine the billable count. After sample preparation, patchCount is the authoritative retained-position count for the sample’s persisted MPP and priced model configuration. A dry run returns that exact count and the credit cost for the request you are about to submit. If the tissue model fails, preparation remains retryable and no price is published. Counting and inference never substitute a different RGB heuristic. There is no separate inference fee, surcharge, or per-job overhead. Only marker predictions cost credits at all. Uploads, ingest, sample and marker reads, dry runs, result downloads, and cell segmentation are free.

​
Getting the exact price before you submit

POST /predict with dryRun: true validates and prices the exact request without charging anything or creating a job. Its patchCount is the exact number of retained MPP-normalized model-grid positions the run will cover, and estimatedCredits is what you will be charged:

​
When you are charged, and when you get credits back

Correcting a sample’s MPP before submission rebuilds its prepared geometry and can change patchCount, so price again after an MPP correction. Submitting a run deducts its full estimated cost from your organization’s balance in one step, before inference starts. The job atomically snapshots the priced MPP, retained position count, model configuration, and prepared artifact identity alongside that reservation. Inference uses this immutable snapshot even if the sample’s MPP is corrected later. The API reports the locked amount as reservedCredits, which means:
  • The price of a submitted run cannot change. If we change our per-patch price later, runs already submitted keep the price you were charged.
  • Failed and cancelled runs are credited back automatically. If a job ends as failed or you cancel it, the full amount returns to your balance. Marker channels already written stay on the sample.
  • You pay only for delivered markers. If a run delivers some of its markers and not others, you are charged for the delivered ones and credited back for the rest.
  • A run you cannot afford is rejected, not partially charged. The request returns 402 with a required field showing the full cost.

​
Buying credits

Credits are sold in prepaid tiers. There is no self-serve top-up yet, so email support@strandai.com with the volume you need and we’ll add them to your organization.

​
See also

  • API reference: POST /predict dry-run pricing, submission, and 402 / 429 behavior.
  • Supported markers: which markers count toward a job’s credit cost.