Phrasly Debuts Ultra Model to Bypass AI Detection

The pressure to evade AI detectors often forces writers to sacrifice the accuracy of their arguments. Phrasly’s internal testing suggests this trade-off is no longer necessary. In a September 2026 benchmark of 72 AI-generated texts, Ultra outperformed three competitors by maintaining a near-perfect human-like score on Pangram, GPTZero, ZeroGPT, and Originality.ai. Unlike rival tools that frequently hallucinate or alter facts during the rewriting process, Ultra averaged fewer than one factual change per document, compared to as many as six in competing models.

CEO Victor Rijo emphasized that the industry has overlooked the importance of argument retention in favor of simple evasion. To address this, the company published their full benchmark data to invite external verification. The tool functions via a five-tier slider—ranging from Gentle to Max—allowing users to calibrate the intensity of edits based on their specific needs. Ultra is now integrated into all paid Phrasly plans, providing a proprietary alternative to standard API-based rewriting services for the company’s base of 5 million global users.