Compare AI humanizers
Pick two to four entries and inspect them side by side. All numbers come from the same shared source set and measurement profiles. An asterisk marks detector-dependent, provisional values.
| November 2026 | Smaug HumanizerVerified by HumanizeBenchWe ran this model ourselves on the benchmark source examples and checked its execution records. Verification confirms the tested model’s provenance; it does not guarantee performance or imply that every detector returned a result.abacusai/Smaug-Humanizer Open weights |
|---|---|
| Shared score components | |
| Overall score | 44.18 |
| Panel bypass score | 18.5% |
| Meaning preservation | 76.5% |
| Raw meaning cosine | 0.9414 |
| Readability | 81.3% |
| Category consistency | 88.2% |
| Detector coverage | 231/231 |
| Identical outputs | 0/33 |
| Penalty points | −1 |
| Per-detector human-likelihood scores · measured outputs only | |
| GPTZero | 4.8% 33/33 |
| ZeroGPT | 0.7% 33/33 |
| Copyleaks | 27.3% 33/33 |
| Originality.ai | 5.5% 33/33 |
| Pangram | 25.6% 33/33 |
| Grammarly | 50.7% 33/33 |
| QuillBot | 99.5% 33/33 |
| Category scores · before entry-level penalties | |
| Academic essay | 44.3 |
| Application essay | 41.0 |
| Blog post | 43.0 |
| Business email | 43.3 |
| Marketing copy | 58.0 |
| Discussion board | 43.0 |
| News article | 38.1 |
| Evidence | All 33 outputs |