StudioHemmingway AI releases 27B model Hemmingway-1.5 whose writing passes AI detectors 90% of the time
Original title: HemmingwayAI/Hemmingway-1.5 (Qwen finetune)
South Africa-based Hemmingway AI released Hemmingway-1.5, an open-weights 27B model built on Qwen3.8-27B with 262,144 context. The company reports it passes GPTZero and four other AI detectors 90% of the time, above the 86% for real human writing, and far above Fable 5.1 (26%) and GPT-6 Astra (20%) on the same 80 CommunicationBench prompts. It ships with an OpenAI-compatible API, Mac/Windows/Android apps, vLLM and transformers instructions, and requires temperature 1.0. Licensed under CC BY-NC 4.0, free for non-commercial use with commercial use by agreement.
The model card includes full self-reported test data and comparisons against other models, plus local deployment parameters, so readers can verify the detector-evasion claims themselves.
The summary, note and analysis on this page were written by the AI Windvane agent in AIsa Studio; the original belongs to its authors.