Primary use stated or inferred cautiously from official documentation.
OpenAI CSAM Image Classifier
OpenAI classifier available since November 16 to help prioritize review of images potentially related to child sexual abuse, including AI-generated material. It combines an internal CLIP visual encoder and an MLP classifier. It neither generates images nor classifies text.
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
Limits and integration
| API ID | omni-moderation-csam-latest |
|---|---|
| Model type | Image classification |
| Access model | Free service for approved organizations |
| License | Proprietary |
| Deployment | Hosted API |
| Release | 16 NOV 2026 |
| Knowledge cutoff | Not published |
| Entrance | Image |
| Exit | Probabilities · JSON |
| Context window | One still image per request; maximum 20,000,000 bytes |
| maximum output | Flags and scores for likely-csam and likely-csam/gen-ai |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Unavailable |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | OpenAI API |
Support specialist trust and safety teams with two risk signals. It complements known-hash detection and human review, with prior approved access and suitable data-handling processes.
A score confirms neither an image’s illegal nature nor its provenance and is not a calibrated probability. The classifier does not report automatically. Known abuse material must not be submitted. Batch API and inline generation moderation are unsupported; quotas depend on the organization.
What it can do
Orientation
Support specialist trust and safety teams with two risk signals. It complements known-hash detection and human review, with prior approved access and suitable data-handling processes.
Context
One still image per request; maximum 20,000,000 bytes · Flags and scores for likely-csam and likely-csam/gen-ai
Tools and integration
Not published
Access
OpenAI API · POST /v1/moderations
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No charge for approved organizations | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | No charge for approved organizations | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
OpenAI states that the service is free for approved organizations. Access must be requested; this is not an unrestricted public API. The unit is one image per request, not LLM generation tokens.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
How to read the results
The card reports 80.2% recall and 99.1% precision for likely-csam at the default threshold, on 6,849 positives and 40,000 negatives. Performance varies by group and domain. These figures must not be used as a guarantee or a general vision evaluation.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| OpenAI CSAM likely-csam offline recall | 80.2% | Recall · threshold 0.996324 | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
Chronology
omni-moderation-csam-latest
Approved access through Moderations API; distinct alias from general moderation.