Qwen3.6-27B Uncensored Release Published

Qwen3.6-27B Uncensored Aggressive has been released. It's based on the original Qwen with refusal responses removed, and is described as a "fully uncensored form of the original release" without any separate personality changes.

The included distributions are as follows.

• Q8_K_P, Q6_K_P, Q5_K_P, Q4_K_P
• IQ4_XS, Q3_K_P, IQ3_M, IQ3_XS, Q2_K_P, IQ2_M
• mmproj for vision support

The author stated that all quantizations were generated using imatrix, and explained that they are compatible with llama.cpp. They also noted that since Qwen3.6 does not support the /think, /no_think soft switches from Qwen3, turning off reasoning mode requires either a Jinja template modification or setting enable_thinking=false.

However, at the beginning of the post, it was noted that a breaking bug had been reproduced in the current release, and that if the issue isn't specific to their own environment, they plan to re-upload the GGUF files. In other words, while the distribution has been posted, it also reveals that this is not a final version whose stability has been fully confirmed.