by OpenAI
OpenAI's free moderation endpoint: the omni-moderation-latest model flags text and images across 13 harm categories, with no per-call charge.
Use the OpenAI Moderation API as a free first filter for text and images in any app, especially one already built on OpenAI. Do not rely on it alone for video, voice or platform-specific rules such as spam: it has none of those. Approved platforms also get a free CSAM image classifier.
Text and image moderation with no per-call charge.
Being free and already inside every OpenAI account makes it the default first layer, while its fixed categories and missing audio and video push serious platforms to add a paid vendor.
Main trade-off
Fixed categories with no audio, video or custom policies.
Free to use with an OpenAI API key, with no per-request charge. Images can be up to 20MB. The separate OpenAI CSAM Image Classifier is also free, for approved organisations only.
The OpenAI Moderation API is a free classification endpoint from OpenAI. Its omni-moderation-latest model scores text and images for harassment, hate, illicit content, self-harm, sexual content and violence, returning a flag and a confidence score for each of 13 categories. It does not accept audio or video.
omni-moderation-latest
One model that classifies both text and images and returns a flag, per-category results and per-category confidence scores.
Thirteen harm categories
Harassment, hate, illicit content, self-harm, sexual content, sexual content involving minors and violence, each with sub-categories such as threats or instructions.
Input-type reporting
Each score says whether it came from the text, the image or both, so mixed posts can be routed correctly.
Inline moderation
The Responses and Chat Completions APIs can return moderation results for the prompt and the reply in the same call that generates the reply.
CSAM Image Classifier
A separate model, omni-moderation-csam-latest, scores single images for potential CSAM and AI-generated CSAM for approved organisations.
SDKs in every major language
Official examples for JavaScript, Python, Go, Java, C#, Ruby and plain curl call the same moderations endpoint.
| Tool | Price from | Free plan | Runs on | Rating |
|---|---|---|---|---|
| Free | Yes | Web | 3.8 | |
| $29/mo | Yes | Web | 3.9 | |
| Quote | No | Web | 3.6 | |
| Quote | No | Web | 3.9 |
Proprietary
Nothing. OpenAI's moderation guide states that the endpoint is free to use, so there is no per-request bill. You need an OpenAI API key to call it.
The docs present it as a general classifier for text and image inputs, not only for checking OpenAI model output, and they show it classifying standalone user content. Check OpenAI's usage policies for your own use case.
Harassment, harassment with threats, hate, hate with threats, illicit, violent illicit, self-harm, self-harm intent, self-harm instructions, sexual, sexual involving minors, violence and graphic violence. Harassment and hate are scored on text only.
No. The general model must not receive child sexual abuse material, known or suspected. Approved organisations can apply for the separate OpenAI CSAM Image Classifier, which returns signals for potential CSAM and potential AI-generated CSAM and does not report to authorities by itself.
No. The current model takes text and images only. For video you would extract frames yourself, and for voice you would transcribe first, or use a vendor with native audio and video moderation.
The OpenAI Moderation API is the content classification endpoint that OpenAI offers to every developer with an API key. Its current model, omni-moderation-latest, reads text and images and returns a flagged verdict plus a score for each harm category. OpenAI does not charge for the endpoint, so it costs nothing to put in front of a paid vendor as a first layer.
A call returns three things: a top-level flagged value, a true or false per category, and a confidence score per category with the input types each score applied to. The categories cover harassment and threats, hate and hateful threats, illicit and violent-illicit content, self-harm, its intent and instructions, sexual content, sexual content involving minors, violence and graphic violence. Some are text-only: harassment and hate, for example, are not scored on images. Developers can also ask for moderation scores inline, in the same request that generates a model reply, through the Responses and Chat Completions APIs.
Child safety is handled apart. OpenAI's docs say never to send known or suspected CSAM to the general model; approved organisations can apply for the OpenAI CSAM Image Classifier, a separate model that is also free and scores an image for potential CSAM and potential AI-generated CSAM. Moderation platforms such as Checkstep have begun offering that classifier inside their own queues.
It is a classifier, not a moderation system. It has no audio or video input, no custom categories, no review queue and no appeals, and OpenAI warns that scores can be recalibrated over time, so thresholds built on them need re-testing. Platforms that need video, live streams or audio should compare Sightengine or Deploy Content Safety, and teams that need the workflow around the scores should look at platforms in AI governance tools.
Product facts on this page (pricing, plans, integrations, compliance) are read from OpenAI's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.