The Cheapest Text Moderation API in 2026: What 1,000,000 Checks Cost on Each
I priced one million text checks a month on six moderation APIs, from each vendor's own pricing page. The bills run from $0 to $1,599.

One million short texts a month. Checked by OpenAI, that costs nothing. Checked by Sightengine at its list rates, it costs $1,599.
Same million comments, same month. I didn’t expect the gap to be that wide, so I went through six moderation APIs and priced each one from its own pricing page, read on November 10, 2026. (AI helped me write this; every price is from the vendor’s own page.)
If you’re picking one right now, maybe because Perspective is going away, this is the whole table, with what each vendor counts as one check. Because that part changes the bill more than the headline price does.
How I counted
Three volumes: 10,000, 100,000 and 1,000,000 checks a month. That’s roughly a small community, a busy app, and a product with real traffic.
Every text is short, under 100 characters, like a chat line, a username or a one-line comment. Long texts cost more on two of these, and I’ll show where.
Prices are list prices in US dollars, per month, before tax, on the cheapest plan that covers the volume. No sales calls, no negotiated rates.
The results

Here’s each one, cheapest first.
OpenAI Moderation: $0 at every volume
OpenAI’s moderation guide says it plainly: “The moderation endpoint is free to use.”
The catch is what it looks for. Its categories are harassment, hate, illicit, self-harm, sexual and violence, with their subcategories, 13 in all. None of them is profanity. So it’s a great free check for abuse and threats, and the wrong tool if you need to stop someone typing a swear word into a username.
Perspective API: $0, until December 31
Perspective’s FAQ also says it’s free, with a default quota of an average of one query per second per project. One million checks a month is about 0.4 a second, so it fits.
But the service ends after December 31, 2026, and Google stopped taking quota requests in February 2026. So the honest price is $0 for seven more weeks, then a migration.
Amazon Comprehend: $3, $30, $300
Comprehend’s toxicity detection is billed in units of 100 characters, with a minimum of three units (300 characters) per request. At $0.0001 a unit, every short text costs $0.0003.
Two things to know before you budget:
- The free tier (50,000 units a month for 12 months) lists eligible APIs, and toxicity detection isn’t on that list.
- Its own worked example for toxicity charges a 100-character comment one unit, not three. At one unit, a million checks would be $100. I’d budget for the minimum the rules state and ask AWS which one applies.
It reads English only, and you can send up to 10 strings per request, each up to 1 KB. It does have a PROFANITY category, next to insults, hate speech, harassment, sexual content, threats and graphic language.
Azure AI Content Safety: $3.80, $38, $380
Azure charges $0.38 per 1,000 text records on its Standard tier. A record is up to 1,000 characters, so a 7,500-character post counts as 8 records. For short texts, one text is one record.
There’s a free tier with 5,000 text records a month, which is enough to build and test against, but not enough for even the smallest volume here.
One thing to know: Azure files profanity under hate, at a low severity, when it isn’t aimed at a group. So you’ll be tuning a severity threshold, not reading a profanity flag.
Hive: $5, $50, $500 (on paper)
Hive’s pricing link now opens deploy.com/pricing, where Text Moderation is $0.50 per 1,000 requests. That’s cheap.
But the self-serve plan says “Limit: 100 requests/day.” That’s about 3,000 a month, under even the 10,000 row. For more, it’s “Contact Sales.” So the $5, $50 and $500 are what the list rate works out to, and none of them is something you can sign up for and run today.
Sightengine: $29, $189, $1,599
Sightengine counts operations, and its docs show a rule-based text check using one. If you turn on its machine learning models, the example with two models uses two, so the bill can double.
Its plans come in steps:
- 10,000: Starter, $29 a month.
- 100,000: Starter’s 50,000-operation step at $99, plus 50,000 more at $0.0018 each, is $189. That beat every other mix of plans and extra operations.
- 1,000,000: Pro’s 200,000-operation step at $399, plus 800,000 more at $0.0015 each, is $1,599. Past 500,000 the page says to contact them, so a quote at this volume may differ.
The free plan has 2,000 operations a month, at most 500 a day.
What the numbers don’t show
What each one catches. Price per check means nothing if the check misses what you care about. OpenAI and Azure are built around abuse and harm. Comprehend has a profanity label but reads only English. If your problem is swearing and the ways people spell around a filter, the cheapest rows may not be the cheapest fix.
Long texts. Everything here assumed short texts. A 2,000-character forum post is two records on Azure and 20 units on Comprehend, while the pages I read for OpenAI, Hive and Sightengine price per request or operation and say nothing about length.
Rate limits. Perspective’s one query a second is an average. Hive’s free limit is per day. If your traffic comes in bursts, the plan that’s cheapest on paper can be the one that throttles you.
Your time. Azure and Comprehend come with a cloud account, IAM and a region. That’s nothing if you’re already there and a whole afternoon if you’re not.
Full disclosure, I built Unprofane, a profanity filter API, so here’s mine in one sentence: $9 a month for 25,000 checks, $29 for 100,000 and $249 for 1,000,000, with free test keys at 100 checks a month and the rest on the pricing page. It costs more than the cloud APIs and less than Sightengine, and what the difference buys is a filter built for profanity and its disguises, in 17 languages.
So which one is cheapest?
Here’s how I’d pick:
- You need abuse and threats caught, and nothing else: OpenAI Moderation. It’s free, and nothing on this list beats free.
- You’re on Perspective today: budget for the move now. Its price is $0 until it’s gone.
- You’re already on AWS and your users write in English: Comprehend, at $300 a million, once AWS confirms the per-request minimum.
- You’re already on Azure: Content Safety, at $380 a million, and plan to tune its severity levels.
- You need high volume from Hive or Sightengine: talk to their sales teams, because the self-serve prices stop well before a million.
And whichever you pick, count your texts’ real length before you trust any row in that table.
Hope this saves you a spreadsheet :)
I’m building Unprofane, an API that catches swearing however it’s spelled.
Catch what Perspective misses. Change one line.
A profanity and toxicity API your users can’t spell their way around, on Perspective’s own path and JSON. Point your client at it and your queue gets better, not quieter.
- 89.0%
- of toxic texts caught at a 7.0% false-alarm rate. Perspective, at the same rate: 58.4%.
- 0 of 549
- innocent words and names flagged on the Scunthorpe Test’s public split: Scunthorpe, cockatoo, Dickinson.
How it was measured, on 9,029 held-back texts.
Hobby50% off for 3 months
$4.50a month for 3 months, then $9
$9a month
25,000 checks a month, every language and disguise included.
Get your API key- Free test keys, 100 checks a month, no card
- Same path, same JSON as Perspective
- Every plan and its price

