OpenAIOpenAI moderation
omni-moderation
omni-moderation classifies potentially harmful text and images. It is a dedicated moderation model, rather than a conversational assistant that writes an answer.
A free classifier has a dated snapshot
OpenAI provides moderation models without a model-use charge. The latest alias currently points to omni-moderation-2024-09-26. The endpoint accepts images as well as text, while the application decides what to do with the returned classification.omni-moderation official model documentation
Keep detection separate from enforcement
Practical guidance. Define how a classification leads to review, a warning or another product action. Review uncertain and context-sensitive cases with people. A model category does not by itself describe the user intent or establish the correct policy outcome.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | API |
| Model identifiers | omni-moderation-latestomni-moderation-2024-09-26 |
| Input and output | text, image in; text outomni-moderation official model documentation |
| API routes | Batch (v1/batch); Moderation (v1/moderations)omni-moderation official model documentation |