Skip to main content

​
Direct answer

Confirm moderation model and endpoint availability in the catalog and console. Do not describe moderation as free, 100% accurate, or universally available. A moderation result is a risk signal, not a replacement for business rules, human review, or legal judgment. This page was last verified on September 2, 2026. Use OpenAI’s current model catalog for the official moderation model and use the LaoZhang API console and call logs for actual models, groups, and billing.

​
Minimal request

When the console explicitly shows an OpenAI Moderations-compatible endpoint:
The example model appears in the current catalog, but permission, fields, and price still require a real check.

​
Handle results

Moderation models can return an overall flag, categories, category scores, or provider-specific fields. Applications should:
  1. store the moderation model and version;
  2. map only categories used by the business policy;
  3. define a human-review band instead of relying on one boolean;
  4. track false positives, false negatives, language, and region differences;
  5. re-evaluate thresholds after a model change.

​
Production boundaries

  • Moderation models can produce false positives and false negatives.
  • Category definitions and scores from different providers are not directly comparable.
  • A moderation result is not a legal or compliance conclusion.
  • Use a safe default and human escalation when moderation fails.
  • Billing, rate limits, and data handling follow the console, call logs, and Data Policy.