About
Language deserves better benchmarks.
Humanizer Arena independently tests AI humanizers with a transparent, reproducible methodology — and publishes the evidence behind every ranking.
Who's behind it
Terry Smith
Terry Smith is passionate about writing and language, especially the choices that make prose clear, natural and distinctive.
He created Humanizer Arena to test AI humanizers on more than detector bypass alone: whether they keep the original meaning, read naturally and produce consistent results across many different texts.
Current benchmark
October 2026 by the numbers
The current cycle reports aggregate results for 10 AI humanizers across five AI detectors.
Current results are aggregate, point-in-time measurements. September remains available as an archived cycle.
What we measure
Four dimensions, one honest score
Detector bypass matters, but it is never treated as success if meaning, readability or consistency suffers.
Detection bypass
How consistently each rewrite avoided a panel of independent AI detectors.
Meaning
Whether the humanized text preserved the original meaning, claims and intent.
Readability
How naturally the output reads to a human, without awkward phrasing or filler.
Consistency
Whether quality held up across many samples rather than a lucky single run.
Research & data
Inspect the evidence
Rankings are produced from identical inputs and fixed scoring criteria, then published with the methodology and supporting data. Results describe measured performance in a specific cycle — not a guarantee for any individual document.
- Current resultsOctober 2026 Humanizer Arena leaderboard
- MethodologyHow every humanizer is tested and scored
- Academia.eduBest AI Humanizers of 2026: A Comparative Benchmark of Detection Bypass, Meaning Preservation, Readability and Consistency
- ZenodoAI Humanizer Benchmark 2026: A Comparative Evaluation of Detection Bypass, Meaning Preservation, Readability, and Consistency
- GitHubAIhumanizerarena/aihumanizerbenchmark
- KaggleBest AI Humanizer — dataset