September 2026 review · GPTZero 2026-09-13-base
Best AI humanizer for GPTZero
In the September 2026 HumanizerEval review, WriteHuman had the best GPTZero 2026-09-13-base result, passing 100% of the 49 texts it returned. All 10 tools rewrote the same 50 English texts, and every output went to every detector in the panel. Review our full results for the best AI humanizer on the leaderboard page.
GPTZero results, September 2026
| Humanizer | Strict bypass | Texts passed | Mean human score | Median human score |
|---|---|---|---|---|
| WriteHuman | 100% | 49 of 49 | 100.0% | 100% |
| StealthGPT Super | 98% | 49 of 50 | 99.0% | 100% |
| Undetectable.ai | 96% | 48 of 50 | 94.7% | 100% |
| Grubby | 94% | 47 of 50 | 93.6% | 100% |
| Rephrasy | 88% | 44 of 50 | 87.3% | 100% |
| RewriteAI | 86% | 43 of 50 | 83.3% | 100% |
| HIX Bypass | 68% | 34 of 50 | 63.6% | 94.97% |
| Humbot | 62% | 31 of 50 | 60.8% | 83.18% |
| BypassGPT | 56% | 28 of 50 | 57.4% | 67.06% |
| Ryne | 8% | 4 of 50 | 8.3% | 0% |
What is GPTZero?
GPTZero launched in January 2023, built by Edward Tian, then a Princeton undergraduate, and Alex Cui, who left a University of Toronto doctorate to co-found it. It raised a $10 million Series A led by Footwork in June 2024, and announced in June 2026 that it was joining Superhuman, the productivity company that also owns Grammarly.
GPTZero describes its detector as a proprietary deep-learning model that takes hundreds of factors into consideration, with perplexity and burstiness as its original public signals, and says it prioritises minimising false positives.
In the Jabarian and Imas study, GPTZero’s false positive rate was under 1% and its false negative rate roughly 0 to 2% on the paper’s corpus.
Who uses it. GPTZero claims more than 17 million users, over 1 million educators and use in over 3,500 colleges, and signed a partnership with the American Federation of Teachers in October 2023.
How GPTZero compares with the rest of the panel
Across the 10 tools tested in September 2026, the average strict-bypass rate on GPTZero was 76%, against 32% on Pangram, 76% on Originality.ai, 95% on ZeroGPT and 92% on Winston AI, on the same texts.
Sources
- How GPTZero works
- Jabarian & Imas, NBER working paper 34223
- HumanizerEval September 2026 review archive: frozen results and downloads
GPTZero FAQ
Which AI humanizer is best for GPTZero?
In September 2026, WriteHuman had the best GPTZero 2026-09-13-base result, passing 100% of the 49 texts it returned. The next best was StealthGPT Super, passing 49 of 50 texts (98%).
Does StealthGPT Super bypass GPTZero?
Yes. In September 2026, StealthGPT Super passed GPTZero 2026-09-13-base on 49 of 50 texts (98%). Its mean human score was 99.0% and its median was 100%.
Does Undetectable.ai bypass GPTZero?
Yes. In September 2026, Undetectable.ai passed GPTZero 2026-09-13-base on 48 of 50 texts (96%). Its mean human score was 94.7% and its median was 100%.
Does Grubby bypass GPTZero?
Yes. In September 2026, Grubby passed GPTZero 2026-09-13-base on 47 of 50 texts (94%). Its mean human score was 93.6% and its median was 100%.
Does Rephrasy bypass GPTZero?
No. In September 2026, Rephrasy passed GPTZero 2026-09-13-base on 44 of 50 texts (88%). Its mean human score was 87.3% and its median was 100%.
Does RewriteAI bypass GPTZero?
No. In September 2026, RewriteAI passed GPTZero 2026-09-13-base on 43 of 50 texts (86%). Its mean human score was 83.3% and its median was 100%.
Does HIX Bypass bypass GPTZero?
No. In September 2026, HIX Bypass passed GPTZero 2026-09-13-base on 34 of 50 texts (68%). Its mean human score was 63.6% and its median was 94.97%.
Does Humbot bypass GPTZero?
No. In September 2026, Humbot passed GPTZero 2026-09-13-base on 31 of 50 texts (62%). Its mean human score was 60.8% and its median was 83.18%.
Does BypassGPT bypass GPTZero?
No. In September 2026, BypassGPT passed GPTZero 2026-09-13-base on 28 of 50 texts (56%). Its mean human score was 57.4% and its median was 67.06%.
Does Ryne bypass GPTZero?
No. In September 2026, Ryne passed GPTZero 2026-09-13-base on 4 of 50 texts (8%). Its mean human score was 8.3% and its median was 0%. A median of 0% means more than half of its outputs were flagged as fully AI-written.
Is GPTZero more accurate than the other detectors?
No. HumanizerEval measures how often humanized text passes a detector, not how accurate that detector is on unmodified writing. Those are different questions. On accuracy, the Jabarian and Imas study (NBER working paper 34223) found Pangram near zero on both false positives and false negatives, GPTZero under 1% false positives, and Originality.ai missing 10 to 40% of AI text depending on the generator. Winston AI was not in that study.