omni-moderation-latest logo

omni-moderation-latest

Visit

omni-moderation-latest is OpenAI's free moderation model: it classifies text and images into 13 categories through the /v1/moderations endpoint.

Share:
View alternatives

omni-moderation-latest

omni-moderation-latest is the alias for OpenAI's current multimodal moderation model, served by the Moderations endpoint. According to OpenAI's moderation guide, the moderation endpoint is free to use, accepts text and image inputs, and returns flags and category scores you can use to filter content. Image files can be up to 20 MB, and audio is not supported.

Status

Live. The guide lists no deprecation. It does say the underlying model will be upgraded over time, so custom policies built on category_scores may need recalibration after an upgrade. Because the name is an alias, behavior can change without a code change on your side.

Key Features

  • Free: the guide says the moderation endpoint is free to use.
  • Multimodal: text and image inputs in one request.
  • 13 categories: harassment, harassment/threatening, hate, hate/threatening, illicit, illicit/violent, self-harm, self-harm/intent, self-harm/instructions, sexual, sexual/minors, violence and violence/graphic.
  • Per-category input support: the guide notes that some categories are text only (harassment, harassment/threatening, hate, hate/threatening, illicit, illicit/violent and sexual/minors), while self-harm, sexual, violence and violence/graphic support text and images.
  • Scores and flags: use the boolean flags directly, or set your own thresholds on category_scores.
  • Endpoint: POST https://api.openai.com/v1/moderations. The guide also describes requesting moderation inline with the Responses API through a moderation parameter.

The guide does not specify which languages are supported.

Use Cases

  • Screening user prompts and uploaded images before they reach your application or another model.
  • Checking model outputs before showing them to users.
  • Triage queues for human review, ranked by category score.

Limitation: the categories are fixed. If you need custom policies, consider a policy-adaptive classifier such as Mistral Shieldstral 1.0.

Pricing

Free to use, according to the moderation guide.

Getting Started

  1. Send a POST request to /v1/moderations with model set to omni-moderation-latest and your text or image input.
  2. Read flagged and category_scores from the response.
  3. Re-test your thresholds when OpenAI upgrades the model.

Frequently Asked Questions

Is the OpenAI moderation API free?

Yes. The moderation guide says the endpoint is free to use.

Does it moderate images?

Yes. It accepts text and image inputs, with images up to 20 MB. Audio is not supported.

What categories does it detect?

Thirteen: harassment, harassment/threatening, hate, hate/threatening, illicit, illicit/violent, self-harm, self-harm/intent, self-harm/instructions, sexual, sexual/minors, violence and violence/graphic.

Will the model change?

OpenAI says the underlying model will be upgraded over time, so recalibrate any custom thresholds.

Alternatives

Conclusion

omni-moderation-latest is a free, fixed-category moderation endpoint for text and images. Use it as a low-cost first filter, and pin your own thresholds against the official guide since the underlying model will change.

Comments

No comments yet. Be the first to comment!