# AI Humanizer Benchmark

> A monthly measured benchmark of AI humanizers. Every tool rewrites the same 33 freshly generated texts on its default settings; every output is scored by 7 commercial AI detectors (GPTZero, Winston AI, Originality.ai, ZeroGPT, Grammarly, QuillBot, Copyleaks) plus meaning preservation and readability. All inputs, outputs, detector verdicts, and the scoring code are published, so every ranking can be recomputed independently.

Current cycle: September 2026, run September 3, 2026. Full ranking:

1. UndetectedGPT — 84.9/100
2. SmartHumanizer — 84.8/100
3. WriteHuman — 81.6/100
4. GPTinf — 78.3/100
5. AI Humanize — 76.1/100
6. HIX Bypass — 73.8/100
7. StealthGPT — 71.9/100
8. SuperHumanizer — 68.5/100
9. CleverHumanizer — 66.8/100
10. ReHumanize — 39.0/100
11. Undetectable AI — 38.2/100

Scores are a 0–100 composite: detector bypass 42%, meaning preservation 32%, readability 16%, consistency across writing categories 10%, minus penalties for quality failures. Prompts are fixed cryptographically before each run (commit–reveal), published cycles are never edited, and no ranking position can be bought. Some ranked tools are made by the benchmark's operator; they run through the identical pipeline, and the published raw data makes any special treatment detectable.

## Rankings

- [Leaderboard](https://aihumanizerbenchmark.com/leaderboard): all tools, all scores, downloadable per-tool data
- [Compare](https://aihumanizerbenchmark.com/compare): side-by-side comparison and every head-to-head matchup
- [Best-for rankings](https://aihumanizerbenchmark.com/best-for): reweighted for essays, students, SEO, marketing, specific detectors, and free tools

## Per tool and per detector

- [Humanizers](https://aihumanizerbenchmark.com/humanizers): one page per tool with per-detector and per-category breakdowns and the full run-by-run evidence table
- [Detectors](https://aihumanizerbenchmark.com/detectors): the panel ordered by strictness, with a bypass ranking per detector

## Method and proof

- [Methodology](https://aihumanizerbenchmark.com/methodology): the complete rulebook, formula, penalties, and how to recompute any score
- [Evidence](https://aihumanizerbenchmark.com/evidence): every prompt with every tool's rewrite and every detector verdict
- [Why this exists](https://aihumanizerbenchmark.com/why): the problem with existing humanizer reviews, and how conflicts are handled
- [Raw data and verification](https://github.com/AIHumanizerBenchmark/AIHumanizerBenchmark): per-cycle bundles (CC BY 4.0) and a zero-dependency verifier (MIT); `npm run verify` recomputes everything

## Full detail for LLMs

- [llms-full.txt](https://aihumanizerbenchmark.com/llms-full.txt): the complete methodology and every published number for the current cycle in one plain-text document

## Updates

- [Blog](https://aihumanizerbenchmark.com/blog): cycle recaps and methodology notes
- [RSS](https://aihumanizerbenchmark.com/rss.xml)