Can Turnitin detect Claude 5?

What Claude 5 is, and why students are asking

Claude 5 is Anthropic’s newest model family. The first generally available model, Claude Fable 5, launched in June 2026 and is positioned above the Opus models that came before it. It writes long-form prose noticeably more naturally than earlier models, which is exactly why the question comes up every time a new model ships: does the detector know about it yet?

The question is reasonable. The assumption behind it, that detection works like a list of known models, is not.

How Turnitin treats a brand-new model

Turnitin’s AI writing detection model does not look up which tool produced your text. It splits long-form prose into overlapping segments of a few sentences, scores each segment for machine-writing patterns, and only reports a document score when enough qualifying sentences cross its threshold. It needs roughly 300 words of prose to run at all, and scores under 20% are hidden behind an asterisk instead of shown as a number.

That design means there is no "Claude 5 entry" that has to be added before Claude 5 text can be flagged. Any sufficiently machine-like prose can trigger it, whichever model wrote it. For the full model-level breakdown, read what AI detector Turnitin uses, and for the basics, does Turnitin detect AI.

Does a newer, more natural model beat the detector?

Sometimes partially, never reliably. Two things are true at once. A brand-new model can sit in a lag window where detectors score its output less confidently, because the detector was trained mostly on older models’ text. And detectors keep catching new models anyway, because large language models share statistical habits that survive version bumps: even sentence rhythm, safe word choices, tidy transitions, and low-surprise phrasing.

Turnitin has also retrained before. Its first detection model was replaced within a year of launch, and a separate model for AI-paraphrased text was added after that. So even if a gap exists on launch day, it is rented, not owned. Anyone quoting exact Claude 5 pass rates this soon after release is guessing; treat those claims with suspicion in both directions.

The practical problem is asymmetry. You cannot measure the lag window from outside, but you carry the whole downside if your paper is the one that gets flagged.

We test with real reports, not guesses

PaperCheck runs documents through the real Turnitin and returns the official AI Writing and Similarity Reports. We are running Claude 5 samples through the actual detector as checks come in, and we will update this post with real scores as results accumulate, the same way we did for ChatGPT detection.

If you used Claude 5 in your draft

Treat the output as notes, not final prose. Rewrite from your own outline, put your course concepts and sources back in, vary your sentence rhythm, and keep drafts that show how the argument developed. A paper you can explain out loud is a paper you can defend.

Do not chase humanizer tools to scrub the text; that trades one detection risk for another. And if genuinely human writing still gets flagged, follow the steps in the Turnitin false positive guide.

How to check Claude 5 text before submitting

Start with the free AI estimate for a fast signal on which passages still read machine-like. If the assignment is high stakes, run the real Turnitin check so you see the exact AI and similarity numbers your instructor would see, before they do. For what the percentage itself means, read the Turnitin AI score explained.

Frequently asked questions