StealthGPT vs Winston AI
StealthGPT Super passed 48 of 50 checks (96%) on Winston AI 5.2 in HumanizerEval’s November 2026 benchmark. Compare all 15 humanizers and learn how the study used Winston’s Human Score.
HumanizerEval · November 2026
StealthGPT Super vs Winston AI 5.2
StealthGPT Super passed 48 of 50 checks (96%) on Winston AI 5.2.
Eight configurations reached 50 passes on Winston AI. The same StealthGPT Super outputs passed 30 GPTZero checks and 19 Originality.ai 5.0 checks under those detectors’ rules.
The StealthGPT team completed this HumanizerEval run on November 17, 2026. Each of the 15 services rewrote the same 50 English inputs. StealthGPT used Super with Quality mode and rephrase enabled.
A document Human Score of 80 or higher qualifies. This detector contributes 5% of the weighted bypass score. Each table rate shows the share of tested outputs that met the pass rule.
| Humanizer | Passes | Rate |
|---|---|---|
| Phrasly Ultra | 50 / 50 | 100% |
| WriteHuman | 50 / 50 | 100% |
| Undetectable.ai | 50 / 50 | 100% |
| Clever Humanizer | 50 / 50 | 100% |
| ZeroGPT Humanizer | 50 / 50 | 100% |
| BypassGPT | 50 / 50 | 100% |
| Grubby | 50 / 50 | 100% |
| Humbot | 50 / 50 | 100% |
| StealthGPT Super | 48 / 50 | 96% |
| RewriteAI | 48 / 50 | 96% |
| HIX Bypass | 46 / 50 | 92% |
| Rephrasy | 44 / 50 | 88% |
| Solvely | 44 / 50 | 88% |
| Grammarly Pro | 36 / 50 | 72% |
| QuillBot | 15 / 50 | 30% |
How Winston AI scores a document
The Winston AI text detector reports a Human Score for the whole document and a prediction map for shorter passages. Start with the document result and use the highlights to see which parts of the text contributed to it. The score expresses the detector’s assessment of the text, and the highlighted passages give you context for an editorial review.
HumanizerEval applied an 80-point document threshold for Winston. Pangram, GPTZero, Originality.ai 5.0 and ZeroGPT used the detectors’ qualifying human verdicts. The rates tell you how often the rewritten texts cleared each rule on this sample.
Winston AI benchmark questions
StealthGPT Super passed 48 of 50 checks (96%) on Winston AI 5.2 in HumanizerEval’s November 2026 benchmark. HumanizerEval counted a document Human Score of 80 or higher as a pass.
Winston uses a 0–100 score to estimate how closely the full document resembles human writing. The AI Prediction Map adds information about individual passages. The Winston AI guide explains how to read the two together.
Yes. HumanizerEval scored the same saved outputs on all five detectors. StealthGPT Super passed 48 of 50 on Winston AI 5.2, 44 on Pangram 4.1, and 30 on GPTZero 5.0b. Each detector had its own pass rule.