Free content moderation APIs in 2026: the limits and the catch

Several moderation APIs can be used without paying, but “free” means different things. OpenAI’s moderation endpoint is free to use within rate limits; Google, Azure and Sightengine give a monthly free allowance; AWS’s page lists 1,000 free images a month for 12 months and, for new AWS customers since 15 July 2025, Free Tier credits; and Perspective takes no new users and shuts down after 31 December 2026. Detectivision is paid, with a free trial. The table gives each limit as read on the vendor’s own page, with the date.

Free content moderation APIs at a glance

Every row comes from the provider’s own pages, read on 16 October 2026. “Not listed” means those pages do not list it, not that it fails. Detectivision is one of the providers; its row comes from our pricing page and docs.

We read these pages again every quarter. Next review: by 14 January 2027.

Three kinds of free: free to use, a free allowance, a free trial

“Free” on these pages means one of three things, and each ends differently.

• Free to use. Calls cost nothing, and a rate limit caps them. OpenAI’s moderation endpoint works this way. Open-source models and libraries are free per call too, but you run them yourself.
• A free allowance. A monthly amount that resets: Google’s 1,000 SafeSearch units and 50,000 moderateText units, Azure’s 5,000 text records and 5,000 images, and Sightengine’s 2,000 operations. AWS’s 1,000 images a month is an allowance that lasts 12 months from account creation. Past the allowance, Google bills the extra, Azure’s free tier stops, and Sightengine’s free tier is capped by the day and by the month.
• A free trial or starting credits. A balance or a period that ends: AWS’s Free Tier credits for new customers, Hive’s “$50+ in Free Credits” after you add a payment method, and Detectivision’s trial.

An allowance can keep a small app going, if its terms allow production use. A trial or credits run out, so know the paid price before you build on them.

OpenAI moderation (omni-moderation-latest)

OpenAI’s guide says: “The moderation endpoint is free to use, and image files can be up to 20 MB.” Its model, omni-moderation-latest, takes text and images; the guide adds that “It doesn’t classify audio”. It also says you can “classify standalone inputs” as well as score a response a model generated.

It checks 13 categories. Seven are text only, among them harassment, hate and illicit content. Six take text and images: the three self-harm categories, sexual, violence and violence/graphic. The guide publishes no list of languages, so test it with your own users’ messages.

What caps it is your account’s usage tier. The model page, read on 16 October 2026, lists these rate limits:

The default snapshot behind omni-moderation-latest is omni-moderation-2024-09-26. Our page on the OpenAI moderation API covers its limits, and when a paid API is worth it.

Google Cloud: Vision SafeSearch and Natural Language moderateText

For moderation, Google Cloud has two free allowances, one for images and one for text.

Vision SafeSearch rates an image on five categories, adult, spoof, medical, violence and racy, as a likelihood from VERY_UNLIKELY to VERY_LIKELY. Each feature applied to an image is a billable unit, and “the first 1000 units used each month are free”. After that, SafeSearch costs $1.50 per 1,000 units up to 5 million a month and $0.60 above, or, in Google’s words, it is “Free with Label Detection”, a feature with its own price on the same page.

Natural Language moderateText returns 16 categories, among them Toxic, Insult, Profanity, Derogatory, Sexual and Violent. The first 50,000 units a month are free. For moderation, a unit is 100 Unicode characters and each request is rounded up to whole units, so the allowance covers 50,000 requests of up to 100 characters each. Above it, units are billed in tiers.

Google lists nine languages as supported for moderateText and five as “limited”, Arabic among them. Turkish is not listed. Google adds: “The API does not have a hard restriction on language support, so unlisted languages will still return predictions.”

Amazon Rekognition: the 12-month free tier, the newer credits and Bulk Image Analysis

AWS’s pricing page shows two free offers side by side. Which one applies depends on your AWS account, so check it there before you count on either.

• The 12-month free tier. “During the free tier period you can analyze 1,000 images per month for free each, in Group 1 and Group 2 APIs.” DetectModerationLabels, the moderation call, is in Group 2, and “the free tier period lasts 12 months from the date of account creation.” Rekognition Video’s free tier adds 60 minutes of video analysis a month for 12 months, content moderation included.
• Credits for new customers. “Starting July 15, 2025, new AWS customers will receive up to $200 in AWS Free Tier credits”, which can be spent on Rekognition. At sign-up they choose a free plan or a paid plan, and “the free plan will be available for 6 months after account creation.” AWS’s Free Tier page adds that such an account “closes on its own 6 months after you open it or when your credits run out, whichever comes first.”

After the free part, Group 2 calls, DetectModerationLabels among them, cost $0.0010 per image for the first million images a month and less above that, per the Group 2 pricing example on AWS’s page.

Bulk Image Analysis and Streaming Video “will no longer be available to new customers effective April 30, 2026”. Accounts that used them within the last 12 months keep access, and AWS says the change “does not impact the availability of other Amazon Rekognition features.” For a batch of images stored in S3, the notice points to the Rekognition Image API instead.

Azure AI Content Safety

Azure’s free tier gives 5,000 text records and 5,000 images a month, at up to 5 requests a second. Azure states: “Usage is stopped if the transaction limit is reached on the Free tier. Customers cannot accrue overages on the free tier.” On the paid tier a text record is up to 1,000 characters, so a 7,500-character text counts as 8 records, in Azure’s own example. Azure’s retail price list gives the Standard tier in East US as $0.375 per 1,000 text records and $0.75 per 1,000 images.

It checks text and images for hate, sexual content, violence and self-harm, with severity levels. Microsoft says its text classifiers “are trained and tested on Chinese, English, French, German, Italian, Japanese, Portuguese, and Spanish”, and that they “can classify content in other languages, but those languages aren't included in the trained-and-tested set and performance can vary.” Its Content Moderator overview says Azure AI Content Safety “supports more than 100 languages”. Turkish and Arabic are not among the eight.

The older Azure Content Moderator is deprecated and “will be retired on March 15, 2027”. Our Azure Content Moderator alternative covers that move.

Sightengine

Sightengine’s free tier gives 2,000 operations a month, at most 500 a day, at one request a second, with no time limit. Its pricing page calls it “Perfect for testing the API and building proof-of-concepts.” The free tier lists “Basic Content Moderation” and “AI Detection”. Text moderation and video are listed from the Starter plan, $29 a month for 10,000 operations, and live-stream moderation on Pro, $99 a month for 40,000. Paid plans charge $0.002 per extra operation.

An operation is, for example, one Nudity Detection on one image, and requests that end in an error are not counted. Its text moderation lists Turkish, both for its rule-based checks and for its text classification models.

Perspective API: existing users only, until it shuts down

Perspective API, from Jigsaw, is sunsetting. Its site says: “The service will remain active until December 31, 2026.” Usage and quota requests were handled until February 2026: “After that, we will no longer accept any new requests.” Jigsaw will not offer “direct migration support”.

So Perspective is not an option for a new project, and current users have until the end of 2026. Our Perspective API migration guide maps its attributes to Detectivision flags.

Open-source models: free per call, not free to run

An open-source classifier costs nothing per call, because you run it. Two examples, as their own pages describe them on 16 October 2026:

• NSFWJS (npm package nsfwjs, MIT licence, version 4.4.0 of 4 August 2026) detects NSFW content “client-side” and sorts an image into five classes: Drawing, Hentai, Neutral, Porn and Sexy.
• Falconsai/nsfw_image_detection is an image-classification model on Hugging Face under the Apache 2.0 licence. It labels an image “normal” or “nsfw”.

Both look only for nudity and sexual content in images; neither checks text, violence or hate.

What you pay instead:

• Hosting. Servers sized for your busiest hour, and a GPU for larger models.
• Upkeep. Updates, monitoring, and someone who looks at the misses.
• Testing. A model card is not a test on your content. Measure the model on your own images or texts first.
• Licences. Code and model weights can carry different licences. Read both.
• Where it runs. A check in the user’s browser is a hint, not a gate: users can skip code that runs on their own device. Check again on your server before you publish.

Our NSFW detection API page covers the hosted route.

Free profanity filters: libraries and APIs

If you only need to catch swear words, there are three free routes: a library in your own code, a small free web service, or a free API allowance that has a profanity category.

Libraries are free and work offline, with no network call. The four that work from a word list share one catch: the list finds the words and spellings on it and misses the rest, and it can flag innocent words that contain a listed one, which is known as the Scunthorpe problem. None of these five pages mentions Turkish.

Free web services exist as well. PurgoMalum calls itself “a simple, free, RESTful web service for filtering and removing content of profanity, obscenity and other unwanted text”. It works from an internal word list plus a list of “safe words” such as “class”; its page states no usage limits or terms, and its latest service update is dated 6 January 2022. API Ninjas’ Profanity Filter API takes up to 1,000 characters a request, and its FAQ says: “Commercial use of the Profanity Filter API is not permitted on the free tier.”

Among the free allowances above, Google’s moderateText has Profanity and Insult categories. Sightengine’s text moderation, which its pricing page lists from the Starter plan, also has a profanity category.

Detectivision’s text check has separate profanity and insult flags, and both come with the trial. Like every Detectivision flag today, they judge the whole message: they do not say which word it was, and they do not mask it. Results are reliable in English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian and Hindi, though the text docs note that a flag may cover only some of them. The profanity filter API page covers both flags in detail. For hate and other harmful messages, see toxicity detection.

Detectivision: a trial, not a free tier

Detectivision is paid: it has no free tier and no plan that stays free. Can you try it without paying? Yes. Basic is free for 2 months, with 10,000 text, 10,000 image and 300 video credits. A trial’s unused credits end with it, under our terms. After the trial, Basic costs $49 a month, with 10,000 text and 10,000 image credits (all plans).

Credits are counted per kind of content. When the result comes by webhook (async requests), text uses a credit per started 2,500 characters and an image checked in detect mode uses a credit; an answer given at once costs twice that. Errors and partial results cost nothing. When a kind’s credits run out, its checks are refused and nothing is charged (credits and billing). The API takes up to 1,200 calls a minute per key (rate limits).

Today it checks text and images. Text in languages other than English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian and Hindi is checked, but the results are not reliable (text moderation). Video, audio and live streams are coming soon, so the video credits a plan shows cannot be used yet.

Start the free trial

What Detectivision does not do

• No free tier: after the trial, checks are paid in credits.
• No video, audio or live streams yet: they are coming soon.
• No reliable results for text outside English, Spanish, Portuguese, French, German, Italian, Turkish, Arabic, Russian and Hindi.
• No positions: flags judge the whole text or image, so current plans return no word positions, boxes, masked text or blurred copies.
• No findings below a flag’s threshold of 0.5: they never come back, so act above your own line (flags).
• No CSAM detection, hash matching or age verification.

How to choose a free moderation API

No free tier is right for every app. Rank them against what you need:

1. Content types. List what your users send: text, images, video, audio. A free text allowance does nothing for photos.
2. Languages. Find your users’ languages on the provider’s own list, then test with real messages. “Not listed” does not mean it fails: Google says unlisted languages still return predictions, and Microsoft says performance in languages outside its tested eight can vary.
3. What happens at the limit. Depending on the provider, a free tier stops, starts billing, refuses calls over a rate limit, or closes the account. The table above and the FAQ give each provider’s answer.
4. Data terms. Read where content goes, how long it is kept and what it may be used for, in each provider’s terms. Ours are in the privacy policy.
5. Production use. Some free tiers are meant for testing, and some exclude commercial use. Read the terms, and know the paid price before launch.

For paid APIs side by side, read content moderation APIs compared. For how per-check pricing behaves as you grow, read pay-per-request content moderation.