StealthGPT passed Originality.ai Turbo on 92% of texts (46 of 50) in HumanizerEval's October 2026 review. On Originality.ai 5.0, the strictest detector in the panel, it passed 34%, while no other humanizer passed more than 6%. We consider the newer Originality.ai 5.0 model highly aggressive, but would not call it highly reliable without an independent study of its accuracy and false-positive rate. Our 2023 investigation of Originality.ai documented reliability concerns and human-written text incorrectly flagged as AI by an earlier model.
The detector
Originality.ai is an AI detection and plagiarism checking tool launched in 2022, built specifically for content publishers, SEO agencies, and digital marketing teams. It uses machine learning to assess whether text involves AI, and includes plagiarism detection alongside AI scoring in a single scan. Originality.ai operates on a credits based pricing model. Aggressive detection does not establish reliability: an independent study of the current model needs to assess both AI detection and false positives on verified human writing.
Benchmark
StealthGPT ranked #1 of 11 humanizers in HumanizerEval's October 2026 review.
HumanizerEval ran 11 AI humanizers, including StealthGPT Super, on the same 50 English texts, 364 to 997 words each. Every output was scored through Originality.ai's API using its turbo model. An output counts as a pass only when Originality.ai's overall verdict is human, meaning a human score above 50%; mixed and uncertain verdicts count as fails. The review ran on October 22, 2026, and every per-text score is published in a public CSV.
StealthGPT passed Originality.ai Turbo on 46 of 50 texts (92%). The remaining 4 texts were flagged or marked unclear. The average tool in the panel passed Originality.ai Turbo on 77% of texts.
Originality.ai pass rate, all 11 humanizers, October 2026
| Humanizer | ||
|---|---|---|
| WriteHuman | 100% | |
| Grubby | 100% | |
| Undetectable.ai | 98% | |
| RewriteAI | 98% | |
| StealthGPT Super | 92% | |
| Rephrasy | 86% | |
| Humbot | 76% | |
| Phrasly | 72% | |
| HIX Bypass | 66% | |
| BypassGPT | 56% | |
| Ryne | 0% | |
StealthGPT also passed all 6 detectors on 17 of 50 texts, more than any other tool.
StealthGPT Super passed 46 of 50 texts (92%) on Originality.ai Turbo in HumanizerEval’s October 22, 2026 review.
The review also includes Originality.ai 5.0: StealthGPT Super passed 17 of 50 texts (34%). Explore both model results in the published benchmark.
Originality.ai API · Turbo and model-5.0 settings. The review used 50 English texts per humanizer, 364–997 words each.
Originality.ai’s Bulk Scan accepts documents, spreadsheets and lists of webpage URLs. Its documentation also describes importing URLs from a sitemap where that option is available. For a publisher, this creates a practical review queue: group pages by campaign, content owner or editorial stage before scanning. A well-organized queue makes it easier to connect each report with the article and revision that your team is discussing.
The benchmark on this page includes Originality.ai Turbo and Originality.ai 5.0. StealthGPT Super passed 46 of 50 texts on Turbo and 17 of 50 on 5.0 in the October 2026 review. Use those model names when moving between the table, the published results and your own report. This gives your team a consistent reference when comparing the same article across an editorial workflow.
Yes. Bulk Scan supports collections of documents and URLs, with selectable scan types and downloadable results. A content team can use that workflow to organize an editorial batch rather than opening each item as a separate project.
Originality.ai’s documentation says sitemap import uses the URLs listed in that sitemap. For a complete content inventory, first check which pages your sitemap includes and add any other URLs you want in the batch.
Still have more questions?
Learn more in our support page.
Most of the time. In HumanizerEval's October 2026 review, StealthGPT passed Originality.ai Turbo on 46 of 50 texts (92%) and registered as human written under Originality.ai's AI probability scoring. On the stricter Originality.ai 5.0 model it passed 17 of 50 (34%), the highest rate of the 11 humanizers tested.
Originality.ai combines machine learning based AI detection with plagiarism checking for content publishers and SEO agencies. We consider its newer 5.0 model highly aggressive, but would not call it highly reliable without an independent study of the current model's accuracy and false-positive rate. Our 2023 investigation documented reliability concerns and false positives with an earlier model; those historical findings do not establish the current model's false-positive rate.
StealthGPT's AI humanizer rewrites AI generated content to introduce the structural variation, lexical diversity, and natural unpredictability that distinguish human writing from AI output. StealthGPT's AI humanizer is trained to target the specific signals each major detector uses and is continually updated as those detectors evolve.
HumanizerEval reruns its benchmark every month, and StealthGPT is tested in every review. The latest results, from the October 2026 review, are 92% on Originality.ai Turbo and 34% on Originality.ai 5.0. See HumanizerEval for the newest numbers.
In HumanizerEval's October 2026 review, StealthGPT passed Pangram on 88% of texts, GPTZero, Originality.ai Turbo and Winston AI on 92% each, ZeroGPT on 84% and Originality.ai 5.0 on 34%, for an 80.3% average bypass rate and the #1 rank among 11 humanizers. It passed all six detectors on 17 of 50 texts.
Stop spending hours fixing poorly rephrased AI text. Use StealthGPT and get flawless, human-grade text instantly.
