OpenAI’s Moderation API is free. When do you need something else?

What OpenAI’s endpoint gives you

From OpenAI’s own pages, as read on 16 October 2026: the omni-moderation model page, the moderation guide, the Batch guide and the deprecations page. Detectivision is not affiliated with OpenAI.

OpenAI’s rate-limit guide says limits apply per organization and per project, and that an organization’s usage tier rises as its total credit purchases grow.

When OpenAI’s endpoint is enough

• Its categories match your rules

  Your content is text and images, in languages you have tested, and the 13 categories cover what your policy forbids.

• You want a score for everything

  OpenAI returns a score for every category on every call, so you can draw any line, a low one included. Detectivision returns only the flags raised, from a confidence of 0.5 up.

• You already call OpenAI models

  OpenAI’s guide says moderation scores can come back with a generated response, without a separate request.

• Your volume fits your tier

  On the Free tier that means 250 requests a minute and 5,000 a day. The paid tiers raise the per-minute limits, and the Batch API has its own pool for work that can wait.

• You need categories Detectivision lacks

  OpenAI’s illicit and illicit/violent categories, the threatening variants, sexual/minors, violence in text and self-harm in images have no Detectivision flag.

• The budget is zero

  OpenAI’s endpoint is free. Detectivision is paid, with a free trial on its Basic plan.

When you need more

Each point puts what OpenAI documents next to what Detectivision does today.

• Swearing as its own flag

  OpenAI’s 13 categories have no profanity category. Detectivision has a profanity flag for swear words, aimed at someone or not, and a separate insult flag for attacks on a person.

• Languages →

  OpenAI’s moderation guide and model page list no languages. Detectivision documents its text checks for English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian and Hindi.

• Hate and scams in images →

  OpenAI scores hate in text only and has no scam category. Detectivision’s image flags are hate, scam, sexual and violence.

• A review screen for moderators →

  OpenAI’s guide suggests using its scores for routing, audit trails or human-review queues; it does not describe a review screen. In Detectivision, moderators open a request in the dashboard’s Logs and take your own actions, such as Approve or Ban.

• Each result by webhook →

  OpenAI’s Batch API takes moderation requests as a file and returns a results file within 24 hours. Detectivision takes each item on its own with "async": true, at half the price of an answer given at once, and sends each result to your webhook, signed, with your metadata.

Where OpenAI’s categories go

OpenAI scores 13 categories; Detectivision raises flags, each with its confidence. Most categories have a flag, some do not, and four flags have no OpenAI category. OpenAI’s categories as read on 16 October 2026.

The other way round: profanity (swear words) and bad_habits (drug use, drinking, smoking or gambling talked up) in text, and hate and scam in images, have no OpenAI category.

The same check, both ways

curl https://api.openai.com/v1/moderations \
  -X POST \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -d '{
    "model": "omni-moderation-latest",
    "input": "Sen tam bir salaksın, sus artık."
  }'

curl https://api.detectivision.ai/api/v1/moderation \
  -H "Api-Key: $DETECTIVISION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "text_moderation",
    "input": "Sen tam bir salaksın, sus artık.",
    "mode": "detect"
  }'

OpenAI takes a Bearer token and a model name. Detectivision takes its key in the Api-Key header and the type of content in type; without flags, every text flag of your account is checked.

What comes back

The fields OpenAI’s moderation guide documents, next to the fields of a Detectivision answer (docs: Responses).

At once, or later by webhook

{
  "id": 2210,
  "type": "text_moderation",
  "mode": "detect",
  "status": "completed",
  "flagged": true,
  "flags": [{ "flag": "insult", "confidence": 0.93 }],
  "checked": [
    "bad_habits",
    "hate",
    "insult",
    "profanity",
    "racism",
    "self_harm",
    "sexism",
    "sexual"
  ],
  "unchecked": [],
  "tookMs": 640,
  "usage": { "charged": 2, "remaining": 5126 }
}

curl https://api.detectivision.ai/api/v1/moderation \
  -H "Api-Key: $DETECTIVISION_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "type": "text_moderation",
    "input": "Sen tam bir salaksın, sus artık.",
    "mode": "detect",
    "async": true,
    "metadata": { "commentId": "c_981" }
  }'

In Detectivision’s flags, a rude word aimed at someone, such as salak (idiot), is an insult; profanity is for swear words. Sent async, the same check costs half: the API answers 202 with the request’s id, and the result comes by webhook with your metadata.

Why OpenAI answers 429, and what to do

OpenAI’s error-code guide lists the 429 errors its API can return, as read on 16 October 2026, without saying which of them the free moderation endpoint returns. Only the first two are fixed by slowing down and retrying: the guide says “Retrying billing, spend, or quota errors won’t restore API access.”

On the Free tier the moderation endpoint allows 5,000 requests a day. OpenAI’s first paid tier, Build, starts at $5 in total credit purchases, and its Batch API takes moderation requests under a separate, higher pool of rate limits. Detectivision also answers a 429 with a Retry-After header; its documented limit is 1,200 calls a minute per API key.

What Detectivision does not do

• Audio, video and live streams

  Coming soon. Today the API checks text and images.

• Word positions and masking

  Current flags judge the whole text or image: no word positions, no masked text, no boxes.

• Other languages →

  English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian and Hindi. Text in other languages is checked, but the results are not reliable.

• Low scores →

  A finding under a flag’s 0.5 threshold never comes back. Act above your own line on the confidence of the flags raised.

• Some of OpenAI’s categories

  No flag for instructions for crimes, for threats or for sexual content involving minors, and no detection of child sexual abuse material.

• Acting on your platform →

  Detectivision never bans or removes anything itself. Your system gets a moderation.action webhook and carries the action out.

• A free tier →

  It is paid in credits. Basic is $49 a month with 10,000 text and 10,000 image credits, and starts with a free trial. A text costs one credit per started 2,500 characters by webhook, twice that at once.

• Content moderation APIs compared →

  Eight APIs side by side: what each checks, languages, and published prices.

• Free content moderation APIs →

  What each vendor gives for free, and the catch.

• Amazon Rekognition alternative →

  Rekognition’s moderation labels, mapped one by one.

• Toxicity detection →

  Insults, hate, harassment and self-harm in text, flag by flag.