HemmingwayAI Releases Hemmingway-1.5: A 27B Qwen Fine-Tune for Undetectable AI Writing
HemmingwayAI has released Hemmingway-1.5, a new 27B parameter open-weights fine-tune of the Qwen large language model designed to generate human-like text. The model is specifically optimized to bypass popular AI detectors like GPTZero with high success rates. This release highlights the ongoing efforts in the open-source community to create specialized models that mimic natural human writing styles and evade automated detection tools. Such tools raise important considerations regarding academic integrity, content authenticity, and the reliability of AI detectors. According to the developers, Hemmingway-1.5 passes top AI detectors 90% of the time, outperforming even real human writing samples in bypassing detection. The model weights are freely available for non-commercial use, and quantized GGUF versions are provided for local inference.
## BACKGROUND
Qwen is a family of open-weights large and small language models developed by Alibaba Cloud, frequently used by the open-source community as a foundational base for further fine-tuning. GGUF is a self-contained single file format widely used in the LLM ecosystem to store quantized model weights for efficient local inference on consumer hardware. GPTZero is an automated AI detector designed to identify machine-generated text by analyzing perplexity and burstiness metrics.