The R7-X490 humanizer has been superseded by R10 Stage 1 (Qwen3.5-9B), which scores GPTZero mean 0.537 and 44.2% strict pass on the random40 OOD test, +18.6 points over R7-X490.
Redirecting you now. If nothing happens, click here:
huggingface.co/spaces/erickkwangg/humanizer-r10-stage1-9b
Note: the model is tuned for human voice, not strict faithfulness on fact-dense inputs. If facts matter, treat its output as the first stage of a two-stage pipeline (humanize, then reconcile facts against the original with a general-purpose LLM).