![]()
HumanizerEval Launches the First AI Humanizer Benchmark That Tests Against Pangram
PR Newswire
ST. PETERSBURG, Fla., Nov. 18, 2026
October 2026 leaderboard ranks 11 AI humanizers across six detectors, including Pangram, and names the best AI humanizer
ST. PETERSBURG, Fla., Nov. 18, 2026 /PRNewswire/ — HumanizerEval today launched a public AI humanizer benchmark. The October 2026 review tested 11 humanizers on the same 50 texts against six detectors, including Pangram. The average humanizer passed Pangram on just 17.8% of texts. It is the first benchmark of its kind to include Pangram, one of the toughest detectors humanizers face today. AI humanizers rewrite AI-generated text so detectors are less likely to flag it. The leaderboard is at https://www.humanizereval.com/.
Other humanizer comparisons test tools against GPTZero, Originality AI, ZeroGPT and Winston AI, but none include Pangram. HumanizerEval’s October 2026 average bypass rates were:
|
Detector |
Average bypass rate |
|
Pangram 4.1 |
17.8 % |
|
GPTZero |
46.7 % |
|
Originality AI Turbo |
76.7 % |
|
ZeroGPT |
82.9 % |
|
Winston AI |
88.7 % |
|
Originality.ai 5.0 |
4 % |
A ranking that leaves Pangram out makes nearly every humanizer look better than it is. Per-tool Pangram results are at https://www.humanizereval.com/detectors/pangram/.
“Pangram is the detector that separates a strong humanizer from a weak one,” said Danil Shashkov, XYZ AI Senior Developer. “If a benchmark skips it, its rankings don’t reflect what users actually face.”
Want to Submit Your Humanizer?
If you work for a humanizer company, share your tool and the information we need to review it for a future evaluation. Submissions are not guaranteed placement on HumanizerEval.
How the benchmark works
An output counts as a pass when the detector’s human score is above 50%. Tools are ranked by average bypass rate across all six detectors, and each detector’s results are published separately. Output quality is scored on factual consistency, naturalness, syntax integrity and stance preservation, and never affects the ranking. Methodology, detector versions and tool settings are at https://www.humanizereval.com/methodology/.
October 2026 results
StealthGPT ranked first with an 80.3% average bypass rate. It passed Pangram on 44 of 50 texts, or 88%. Undetectable AI placed second at 71.3%, with the next-best Pangram result at 40%. Humbot and BypassGPT failed Pangram on every text, even though both passed Winston AI every time and ZeroGPT 88% of the time. Originality.ai 5.0 was the hardest detector, and StealthGPT led at 34%.
Open and reproducible
Every figure comes from a dataset of per-text detector scores at https://github.com/StealthGPT-Labs/humanizer-benchmark. Anyone can rebuild the leaderboard with a single script.
About HumanizerEval
HumanizerEval is a public AI humanizer benchmark that ranks AI humanizers by how often their output passes leading AI detectors, including Pangram. It publishes per-detector results, separate quality scores and open data for public verification and research.
Media Contact
Jozef Gherman
HumanizerEval
jozef@stealthgpt.ai
https://www.humanizereval.com/
View original content:https://www.prnewswire.com/news-releases/humanizereval-launches-the-first-ai-humanizer-benchmark-that-tests-against-pangram-302943317.html
SOURCE HumanizerEval
