OpenAI Moderation API

service

proprietary hosted API(no weights, no self-host)

access

free moderation endpoint

source

closed

license

proprietary

Closed, API-only hosted classifier; no weights or source. The documentation URL redirects to developers.openai.com and the page renders client-side, so the reading is taken from the markdown edition of the same document, cited beside it.

• https://platform.openai.com/docs/guides/moderation recorded 2026-08-13

  Still resolves, now via a 301 to developers.openai.com/api/docs/guides/moderation. The page is client-rendered, so the readable body is the markdown edition cited below.

• https://developers.openai.com/api/docs/guides/moderation.md recorded 2026-08-13

  Markdown edition of the cited guide. The product is an endpoint - 'Use OpenAI moderation models to detect harmful content in text and images' via the moderation endpoint or alongside a generated response - with no weights, no download and no self-hostable implementation offered. Use is governed by OpenAI's API terms; 'The moderation endpoint is free to use'.

OpenAI publishes no numbers for this endpoint - no request volume, no developer count, no customer count - and nothing else cited here does either, so no numeric reach is claimed: a 1M-10M label would assert a measurement nobody made. What supports the level is standing rather than a count. This is the free default filter in front of a very large developer base, which is a credible reading but not a counted one. Level 4 on that standing, and nothing in the cited sources argues for a different one.

• https://developers.openai.com/api/docs/guides/moderation.md recorded 2026-08-13

  Guide documents the endpoint and states it is free to use. No usage figure of any kind appears on the page.

Convenient broad-category moderation; fixed policy, no customization vs policy-conditioned models.

• https://developers.openai.com/api/docs/guides/moderation.md recorded 2026-08-13

  omni-moderation-latest 'accepts text and image inputs. It doesn't classify audio.' Thirteen fixed categories - harassment, harassment/threatening, hate, hate/threatening, illicit, illicit/violent, self-harm, self-harm/intent, self-harm/instructions, sexual, sexual/minors, violence, violence/graphic - of which six apply to images and the rest to text only. No mechanism for supplying your own policy.