O

omni-moderation-2024-09-26

Par requête:$0.0016
Usage commercial

Technical Specifications of omni-moderation-2024-09-26

omni-moderation-2024-09-26 is a snapshot moderation model exposed through the OpenAI Moderation API, and OpenAI’s API reference explicitly lists this exact model ID as a supported moderation model option. It is part of the newer omni-moderation family rather than the older text-only moderation line.

From OpenAI’s model and moderation documentation, this model is designed for content safety classification, accepts text and image inputs, and returns moderation results rather than generative text output. OpenAI describes the omni-moderation family as its most capable moderation offering, with multimodal support and broader categorization than legacy moderation models.

As a September 26, 2024 snapshot, omni-moderation-2024-09-26 corresponds to the launch generation of OpenAI’s upgraded multimodal moderation system. OpenAI states that this generation is built on GPT-4o, improves accuracy over prior moderation models, and is especially stronger on non-English content. The same launch materials also note that moderation models are available through the Moderation API at no cost to developers.

What is omni-moderation-2024-09-26?

omni-moderation-2024-09-26 is a fixed-version moderation model used to identify potentially harmful content in application inputs before or after they reach end users. In practice, developers use it to classify user-generated text, images, or mixed multimodal payloads for policy-relevant risk categories such as violence, hate, harassment, sexual content, and self-harm.

Compared with legacy text moderation models, this snapshot belongs to OpenAI’s newer multimodal moderation stack. OpenAI documents that omni-moderation snapshots support more categorization options and multimodal inputs, making them the recommended choice for new moderation integrations.

A key reason to use the snapshot ID omni-moderation-2024-09-26 instead of a moving alias is version stability. Because it is a dated snapshot rather than the rolling omni-moderation-latest alias, it is better suited for teams that want reproducible moderation behavior, controlled rollouts, and more predictable auditing across environments. This stability point is an implementation inference based on OpenAI’s separate listing of snapshot and latest model identifiers.

Main features of omni-moderation-2024-09-26

  • Multimodal moderation: Supports both text and image inputs through the Moderation API, allowing a single model to evaluate content across common user-generated media types.
  • Broader safety categorization: OpenAI documents that the omni-moderation line adds more categorization options than legacy moderation models, including newer harm classes introduced with this generation.
  • Improved multilingual performance: OpenAI states this moderation generation is more accurate than previous models, particularly in non-English and lower-resource language settings.
  • Probability-style scoring output: In addition to flagged category booleans, OpenAI says the model provides calibrated category scores that help teams define application-specific thresholds and review logic.
  • Input-type awareness by category: The moderation response schema includes category-level applied input types, which helps developers understand whether a given classification came from text, image, or both.
  • Snapshot version control: Because omni-moderation-2024-09-26 is an explicitly versioned snapshot, it is useful for regression testing, safety benchmarking, and controlled production deployments where consistency matters. This is an implementation-oriented inference from OpenAI’s snapshot model naming and API support.

How to access and integrate omni-moderation-2024-09-26

Step 1: Sign Up for API Key

To get started, sign up on CometAPI and generate your API key from the dashboard. Once you have an active key, you can use it to authenticate all requests to the omni-moderation-2024-09-26 API through CometAPI’s unified model gateway.

Step 2: Send Requests to omni-moderation-2024-09-26 API

Use CometAPI’s OpenAI-compatible API format and specify the model as omni-moderation-2024-09-26. Send your moderation payload as text, or as a multimodal input array when classifying image-plus-text content. OpenAI’s official Moderation API documentation shows that this model ID is valid for moderation requests and that the endpoint accepts either plain text strings or structured multimodal input objects.

curl https://api.cometapi.com/v1/moderations \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $COMETAPI_API_KEY" \
  -d '{
    "model": "omni-moderation-2024-09-26",
    "input": "Example content to classify"
  }'

Step 3: Retrieve and Verify Results

After sending the request, inspect the response for the top-level moderation result, flagged status, category booleans, and category scores. OpenAI’s moderation response schema also includes per-category details that can be used for thresholding, review queues, and policy enforcement logic. In production, verify results by testing representative edge cases from your own app, comparing outcomes against internal policy expectations, and logging the snapshot model ID omni-moderation-2024-09-26 so moderation behavior remains traceable over time.

Plus de modèles