The best content moderation APIs for 2026 depend on what your users post: a free general-purpose endpoint covers text and images for many apps, while specialists add video, audio, severity scores and human review queues. If you still rely on Google's Perspective API, which Google says ends after 2026, choosing its replacement is this quarter's job, not next year's.
What changed in moderation APIs this year
Pick the wrong moderation layer and you find out in public: a screenshot of something nasty that sailed through, or a wave of angry users whose harmless posts got blocked. This year three shifts make that choice harder to put off.
First, a long-running free option is going away. Google's Perspective API says it's sunsetting and the service officially ends after 2026. If your comment system scores toxicity with it, you need a replacement chosen, wired in and tuned before the switch-off, and at least one text vendor, Tisane, now offers a Perspective migration guide in its documentation.
Second, moderation has become a free utility at the general end. OpenAI says its moderation endpoint is free to use and its latest model reads text and images, though not audio. That changes what paid specialists have to sell: wider coverage, finer categories, review tooling and policy control.
Third, names are moving. Hive now calls itself Deploy (existing customer logins are unchanged), and a vendor that once sold moderation from its own domain now shows a different product there. If a colleague recommends a tool from memory, open its homepage before you open its docs.
Start with the type of tool, then the vendor
Moderation tools come in a handful of kinds, and most teams only need one or two of them. Work out which before you read a single pricing page. For a new feature, my default is the free general endpoint for text and images with a human review queue behind it from day one, adding a specialist only when the policy test further down shows a gap: audio, video, a language your users write in, or severity levels your moderators need.
- Comments and chat on a small app, mostly text: a free general-purpose endpoint as the first pass, with a simple human queue for anything borderline. You don't need a platform yet.
- Replacing a toxicity score: a text API that returns a type and a severity for each flag, so you can map your old thresholds onto new categories instead of guessing.
- Photo uploads on a community or dating app: an image specialist with many moderation classes, plus an optional check on the user's device that stops the obvious stuff before it's uploaded.
- Video, livestreams and voice notes: a vendor that sells video and audio moderation on their own lines. A text-and-image endpoint won't cover audio.
- A trust and safety team with written policies, appeals and regulators to answer to: a platform with review queues and policy tools, not a bare classifier.
- Game or kids' chat full of creative swearing: a profanity filter with lists you can customise, support for several languages and, if your security team insists, an option to host it yourself.
An API is only one layer. Rules, appeals and the people who make the hard calls still sit on top, and this community moderation playbook covers that human side if you're setting it up from scratch.
Ten content moderation APIs worth testing
OpenAI Moderation API
A free classification endpoint for text and images, and an easy first layer for apps that already call OpenAI models.
Azure AI Content Safety
Microsoft's guardrail service for generative AI apps, aimed at teams putting safety checks around their own AI features on Azure.
Hive (now Deploy)
Text and visual models that score how severe a post is rather than matching keywords; the company has renamed itself Deploy.
WebPurify
Profanity filtering, image moderation and human reviewers from one company, for teams that want people checking what the models catch.
Sightengine
Image moderation specialists with a long list of classes, who also cover video, text, audio and spotting AI-generated pictures.
Moderation API
A small vendor bundling text, image, video and audio checks with its own review queue, SDKs and a command-line tool.
Bodyguard
Built around social media comment moderation, with separate APIs for messages, usernames and images that read context, not just words.
Tisane
Every flag this text API raises comes back labelled with an abuse type and a severity, which makes routing to different queues simple.
Cleanspeak
Customisable profanity filtering and machine learning across several languages, offered either as a cloud service or self-hosted on your own servers.
NSFWJS
An open-source JavaScript library that checks images for indecent content on the user's device, before anything reaches your servers.
Want AI doing the heavy lifting in your marketing?
I build the systems that handle the boring 80 percent, so you get your week back. Done properly, with the human kept in.
Mistakes that make moderation worse than having none
Blocking on one score with no review lane. Every classifier gets some calls wrong. If the only outcomes are allow or delete, your users carry the cost of every false alarm. Add a middle outcome, hold for review, and send the uncertain band there.
Moderating text when users post images and voice. Abuse moves to whatever channel you're not checking. If your app accepts uploads or audio, your moderation has to cover them too, even if it's only a lighter check.
Mistaking a classifier's categories for your policy. The categories an API returns are the vendor's labels. Your community rules are yours. Write them down, then map the labels onto them; where nothing maps, you've found a gap you'll need to fill another way.
Leaving the Perspective switch to the last month. Thresholds don't carry over one for one. A score of a given value from a new model doesn't mean what the same number meant from the old one, so plan time to re-tune against real posts.
Trusting a client-side check on its own. Anything that runs on the user's device can be switched off by a determined user. Treat it as a first filter that saves server calls, never as the only one.
Do you make content moderation APIs or trust and safety tools? Tool makers can ask to be considered for a featured place on this list.
Run a policy test before you sign anything
Accuracy figures mean little in the abstract, because moderation is only right or wrong against your own rules. So write the rules, build an answer key, and test two or three candidates against it. It's a day's work, and it'll tell you more than any sales call.
Start by writing your policy as five to eight categories with a one-line definition each: harassment, hate, sexual content, self-harm, spam, plus anything specific to your product, such as off-platform payment requests or personal phone numbers. Then pull two hundred recent items from your own logs, including the ones your moderators argued about, and strip personal data before they go anywhere.
Label each item allow, review or remove. That's your answer key. Send the same set to each candidate on default settings and map their categories onto yours in a spreadsheet. Then count harmful items allowed through, harmless items blocked, and items sent to review. Blocked harmless posts are what your users complain about; the ones that slip through end up in screenshots; and the review pile decides how your moderators spend their week.
Finish with the hard cases: slang, the other languages your users write in, and quoted abuse, where someone reports what was said to them. Keyword filters tend to trip on quoted abuse because the bad words are there, while context-aware models are meant to read who's saying what. That's exactly where a paid specialist has to prove it's worth the fee.
How the bill is worked out
Several specialist vendors list text, image, video and audio moderation as separate products, so your bill depends on the mix of what users post, not just the volume. A text-heavy forum and a video app with the same number of users can end up with very different invoices.
Free options exist, but count them carefully. On the pages read for this guide, OpenAI's endpoint is free to use, Moderation API lists a free Hobby plan, and NSFWJS is open source and runs on the user's device. The other seven don't show a price on the pages checked, so per-item and per-minute rates for video and audio are unpublished there; ask for them in writing, with your monthly mix, before you compare quotes.
Human review is the cost most budgets forget, and you can work it out today. Take a forum with 40,000 posts a month. If the API sends 2% to review, that's 800 items; at ten seconds each, it's a little over two hours of a moderator's time a month. Widen the review band to 5% and it's 2,000 items and five and a half hours. Tuning that band is a staffing decision as much as a safety one.
Want your tool on this list? Featured spots and reviews here are paid partnerships. Here's how to partner with me.

