GPTZero's Updates
Since June 2025, GPTZero has fired off update after update. Five new models in just three months, each one claiming better accuracy, fewer false positives, and newer detection against AI humanizers.
Here’s the rollout:
- June 3 (3.5b) – Detection added for lightly “polished” AI text, stronger bypasser resistance.
- July 23 (3.6b) – Lower false positives, better handling of special characters, tougher on bypassers.
- August 11 (3.7b) – Improved against new LLMs, adversarial prompts, and research data.
- August 22 (3.8b) – Optimized for GPT-5 content.
- September 4 (3.9b) – Across-the-board accuracy boost and stronger recall.
Five updates in under 100 days. That’s not just patching, it’s an AI arms race.
TwainGPT’s Response
Most humanizers don’t last. They work once or twice, then become useless when GPTZero updates again.
TwainGPT isn’t built that way. We constantly update our model to match GPTZero's releases. When they move, we move faster. That’s why TwainGPT stays undetectable.
Real-World Testing
On September 6, 2025, we went head-to-head with GPTZero’s newest model (3.9b):
- Raw GPT-5 essay → detected as 100% AI. Okay, nice work GPTZero.
- TwainGPT rewrite → detected as 1% AI. Better luck next time.
Even in the middle of this arms race, TwainGPT stays ahead.
Conclusion
GPTZero’s updates show how serious they are about AI humanizers. But in this "AI war", TwainGPT keeps winning.
Other tools can’t keep up. And that’s why in 2025, TwainGPT is the only AI humanizer still consistently bypassing GPTZero.