Smaug Humanizer vs. Humanizer CMF
Which open-weight entry preserves source meaning more closely across the shared texts?
Across 33 matched rewrites with numeric meaning preservation scores, Smaug Humanizer scored higher on 23, Humanizer · CMF on 10, and 0 tied. Their November composite scores were 44.18 and 45.23 out of 100, respectively.
Results on the same source texts
Each row compares both entries on the same source and includes pairs with two numeric scores. Meaning uses recorded embedding cosine. Detector scores reflect reported human-likelihood.
| Metric | Smaug Humanizer higher | Humanizer · CMF higher | Tied | Measured pairs |
|---|---|---|---|---|
| Readability | 25 | 4 | 4 | 33 / 33 |
| Meaning preservation | 23 | 10 | 0 | 33 / 33 |
| GPTZero | 2 | 0 | 31 | 33 / 33 |
| ZeroGPT | 1 | 7 | 25 | 33 / 33 |
| Copyleaks | 1 | 6 | 26 | 33 / 33 |
| Originality.ai | 5 | 22 | 6 | 33 / 33 |
| Pangram | 10 | 16 | 7 | 33 / 33 |
| Grammarly | 7 | 21 | 5 | 33 / 33 |
| QuillBot | 6 | 14 | 13 | 33 / 33 |
Read the matched rewrites
These examples show the largest measured gaps in meaning preservation, with one example per writing category.
- Academic essay ? S01
Smaug Humanizer rewrite (0.925) ? Humanizer · CMF rewrite (0.805) - Blog post ? S27
Smaug Humanizer rewrite (0.982) ? Humanizer · CMF rewrite (0.877) - Discussion board ? S21
Smaug Humanizer rewrite (0.920) ? Humanizer · CMF rewrite (0.833)
Scope and source
Both entries use rewrites recorded from identified open-weight artifacts. The comparison uses the recorded outputs in the November 2026 dataset, version 2026-11-v1. See the methodology for metric definitions and limitations.
Model profiles: Smaug Humanizer and Humanizer · CMF.