What the table cannot tell you
Which one is right more often. Not one of these pricing pages states an accuracy rate, a false-positive rate or a precision figure — every percentage on all six is a discount — and that includes ours. So this is a comparison of what things cost, and anybody presenting a price table as a quality ranking is adding a claim the data does not carry.
The number that would actually matter is the one nobody publishes: how often a given tool flags human writing of the kind you produce, at the threshold your institution uses. Until someone reports that, the independent testing is the closest thing available, and what a false positive rate means covers why a single headline accuracy figure hides it.
The vendors that are not in the table
Five, and none of them is missing because it performed badly. Each one is here because its pricing could not be read from the page its server sends.
- GPTZero. Prices render in the browser. The served HTML reads $0/mo throughout, so an automated capture would have recorded it as free. What we do know about it, which is about how it works rather than what it costs.
- Quetext. The page carries a Monthly/Yearly toggle and serves both states interleaved, so the figure in the text is the annual-equivalent with no undiscounted monthly price stated.
- ZeroGPT. No price text served. What we do know about it, which is about how it works rather than what it costs.
- Smodin. Returned nothing on a second request.
- Turnitin. Returns 403 to automated requests, re-confirmed on the capture date. What we do know about it, which is about how it works rather than what it costs.
The GPTZero case is the one worth dwelling on, because it is the trap this whole page was built to avoid. Its pricing page returns a normal 200 response and roughly 1,400 characters of text, in which every price reads $0 /mo — the real figures arrive later, in the browser. An automated capture would have recorded GPTZero as free, published it in a table with a capture date attached, and been confidently wrong in a way that looks sourced. Our method is written up in the methodology, and the rule that saved us here is a dull one: read the figure, do not harvest it.
Why our own row is thin
Because the free plan is the only part of our detector worth putting in a price comparison. It gives 10,000 detection words a month at 1,500 words a scan, it needs an account, and it wins nothing else on this table. Copyleaks and Originality.ai are selling to teams with seat counts and plagiarism checking bundled in; that is a different product to a different buyer.
Where those tools are the better purchase, they are the better purchase. If you need plagiarism matching against a corpus, or seats for a department, this table should send you to one of them.
Method
Each page was fetched with a browser user-agent, stripped of scripts and styles, and the pricing section read from the resulting text. Figures were read rather than pattern-matched, because a regular expression over currency symbols cannot tell a real price from a placeholder, a struck-through comparison price or an unrendered zero. Every price is reproduced with the billing period the vendor stated it with. Where a vendor serves only the annual-billed figure, that is what appears, labelled as such, and the monthly price is not inferred from it.
Nothing on this page is affiliate-linked and no vendor was contacted. We compete with every tool in the table, which is a conflict no disclosure fully cures — the mitigation is that each figure links to the page it was read from, so you can check both the number and how stale it has become.