Tasks and inputs
OpenAI CSAM Image Classifier is listed for Image classification, Content Moderation. It accepts image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
SyncDev model directory · Profile updated Nov 17, 2026
OpenAI CSAM Image Classifier flags potential CSAM and AI-generated CSAM in images to prioritise human review; free for approved organisations.
The specification score summarizes recorded features and limits. It is not a benchmark result.
Pricing conditions: Free of charge for approved organisations as announced by OpenAI; not shown on the public price list. Confirm terms with OpenAI.
Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.
Accepts
Produces
Trust-and-safety teams do not lack signals. They lack time. A review queue with ten thousand images and no ordering is a queue where the worst material waits as long as the harmless. The OpenAI CSAM Image Classifier is designed for the ordering problem: it helps approved organisations prioritise images for human review.
It launched on 16 November 2026 under the model name omni-moderation-csam-latest.
The classifier returns separate signals for two things: potential child sexual abuse material, and potential AI-generated CSAM. Keeping those apart is useful. The escalation path, reporting obligations and investigative follow-up can differ depending on whether an image appears to depict a real child or appears to be synthetic, and a single blended score would hide that distinction.
It is a classifier, not a decision-maker. The output is meant to rank work for trained reviewers, not to replace them.
| Field | Value |
|---|---|
| Model | omni-moderation-csam-latest |
| Released | 16 November 2026 |
| Endpoint | v1/moderations |
| Input | Images |
| Output | Separate classification signals for potential CSAM and potential AI-generated CSAM |
| Price | Free of charge |
| Access | Approved organisations only |
A practical limit: it cannot be used for inline moderation inside Responses or Chat Completions requests. It has to be called on its own, as a standalone moderation request.
There is no charge. Access is limited to approved organisations, and the approval process is handled by OpenAI rather than published as a self-serve form. It does not appear on OpenAI's public price list, which shows only omni-moderation-latest, although OpenAI has announced it as free. Confirm the terms of use with OpenAI when you apply.
Do not route suspected CSAM through the general moderation model. OpenAI's guidance for omni-moderation-latest is explicit: never submit known CSAM to it, and do not send suspected CSAM to it either. The general model covers harassment, hate, self-harm, sexual content, violence and related categories across text and images. This classifier is the separate, access-controlled path for the one category that demands specialist handling.
That gives a sensible pipeline for a platform with both needs:
A model does not discharge your legal duties. Organisations handling this material typically have reporting obligations to the relevant national authorities, evidence-preservation rules and reviewer-welfare policies. OpenAI publishes separate guidance on prevention, detection and reporting practices; read it alongside your counsel's advice before deployment.
You cannot and must not build a test set of real abuse material. Evaluation instead happens within your existing, lawful review operation: run the classifier in shadow mode on the queue your trained reviewers already handle, compare its ordering with their outcomes, and measure how much earlier confirmed cases would have reached a reviewer. Track false positives too, since every one costs reviewer time and wellbeing.
Other safety and moderation models are collected under content moderation, and the full OpenAI catalogue is on the provider page.
Compare the model on the work you need to complete.
OpenAI CSAM Image Classifier is listed for Image classification, Content Moderation. It accepts image and produces text. Use the publisher’s input format and preprocessing for your chosen deployment.
Measure a completed task using the prompts, documents and tools your application will use. Include reasoning and retries in the result.
Use the model identifier and account requirements for the selected API route. Confirm its request limits and output format with one complete task before connecting a larger workflow.
These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.
Mistral AI
Shieldstral 1.0 is a compact open-weight moderation model that classifies text and image inputs against natural-language policy questions.
OpenAI
GPT OSS Safeguard 120B from OpenAI: Safety model for policy screening, moderation, and risk-aware routing workflows.
OpenAI
GPT OSS Safeguard 20B from OpenAI: Safety model for policy screening, moderation, and risk-aware routing workflows.
Meta
Llama Guard 4 12B from Meta: Multimodal Llama 4-based safety classifier for moderating prompts and responses.
Meta
Llama-Guard-3-8B from Meta: Llama 3.1-based safety classifier for moderating prompts and model responses.