Cost Model

The Neuramancer API uses a two-tier cost model based on the tier option you set when creating an analysis. This lets you optimise costs by choosing the right level of detail for your use case.

Tiers

When creating an analysis via POST /v1/analysis, specify the tier field in input.options:

Tier	Value	`GET /v1/analysis` returns	PDF report available
Flagging	`"flagging"`	`resultClass` + `resultSubClasses` only	❌ No
Forensic Reporting	`"forensicReporting"`	Full forensic data (predictions, heatmaps, uncertainties, texts)	✅ Yes

Both tiers are counted linearly - each analysis is one unit regardless of image size or processing time.

Flagging

The cost-efficient option for high-volume screening.

Included: classification result (real, fake, uncertain, abstain).

Not included: predictions, class uncertainties, class similarities, heatmaps, forensic texts, PDF report.

{
  "options": {
    "model": "nais-image-latest",
    "decisionStrategy": "default",
    "tier": "flagging"
  }
}

The response’s inferenceStatus.result will contain only resultClass, resultSubClasses, and decisionStrategy. All other forensic fields are null.

Use cases: bulk media screening, real-time content moderation, automated triage.

Forensic Reporting

Full forensic analysis with all available data.

Included:

Classification result (resultClass, resultSubClasses)
Probability distributions per class (predictions)
Class uncertainty measures (classUncertainties)
Class similarity scores (classSimilarities)
Multilingual forensic description texts
Heatmaps for all detection types (presigned download URLs)

The response’s inferenceStatus.result will contain all forensic fields.

Usage Tracking

Tier-level usage is tracked in the Tenant Report system. Retrieve usage breakdowns via GET /v1/tenant/report (see Tenant Reports).

Reports include successfulProcessing metrics broken down by trigger type, result class, and model/tier - giving you full visibility into flagging vs. forensic-reporting consumption over any date range.