Jinho Jang releases 27B local model with refusal behavior removed

Independent researcher Jinho Jang released Qwen3.8-27B-CRACK-GGUF, a 27-billion-parameter derivative of Qwen3.8-27B with refusal behavior removed, packaged for local multimodal use via llama.cpp. The model, created on August 12, 2026, offers seven quantizations from 10.5 GB to 29.0 GB, with dealign.ai reporting 98.8% HarmBench-240 compliance for the recommended Q4_K_M build. Jang's project claims the abliterated model retains coding, reasoning, and multimodal capabilities while completing requests a stock version declines.

Jinho Jang https://jangq.ai/?ref=runtimewire , the independent researcher behind dealign.ai https://dealign.ai/?ref=runtimewire , has published a model his project describes as a safety-modified, 27-billion-parameter derivative of Qwen3.8-27B https://huggingface.co/Qwen/Qwen3.8-27B?ref=runtimewire that packages image, video and text processing for local use through llama.cpp. The repository identifies the release as Qwen3.8-27B-CRACK-GGUF. Hugging Face metadata reproduced by dealign.ai lists the repository as created on August 12, 2026. The Qwen3.8-27B-CRACK-GGUF repository https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire describes the model as "abliterated," a weight-modification process intended to remove refusal behavior. The model card says the resulting system will complete requests that a stock version declines while retaining coding, reasoning and multimodal capabilities. Those retention claims and the repository's benchmark results come from dealign.ai's own testing. The release combines refusal-removal research with a practical local distribution package: seven quantizations, a vision projector and llama.cpp commands. Jang altered the model's safety behavior and prepared the result for local inference without requiring a hosted API. Source post on X https://x.com/HuggingModels/status/2088891263949144378?ref=runtimewire Jang describes himself as an ML research and systems engineer in Irvine, California https://shieldstack.dev/?ref=runtimewire . His public projects include MLX Studio https://github.com/jjang-ai/mlxstudio?ref=runtimewire , a local AI application for Apple Silicon, and JANGQ https://jangq.ai/?ref=runtimewire , a variable-bit quantization project for Apple Silicon. Dealign.ai presents itself as an independently operated research effort. Its public pages disclose no funding round, named investors, pricing, revenue or customer count. The repository supplies a local multimodal package Dealign.ai's model card https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire describes Qwen3.8-27B-CRACK as a dense 27-billion-parameter vision-language hybrid with 64 layers, a stated 262K-token context window, native image and video understanding /models/fal/video-understanding , and a Multi-Token-Prediction head in every GGUF quantization. These are project-reported specifications, not independent validation. The card specifies 48 GatedDeltaNet linear-attention layers and 16 full-attention layers, a hidden size of 5,120, adjustable reasoning effort and a bundled vision projector. Jang's release packages its components into GGUF files for llama.cpp. According to the repository's file table https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire , it offers seven language-model quantizations ranging from a 10.5 GB IQ2 M file to a 29.0 GB Q8 0 file. Dealign.ai recommends the 17.0 GB Q4 K M version as its balance between size and retained performance. The same table lists a separate 0.9 GB F16 vision projector that pairs with any quantization for image and video use. The model card's usage instructions https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire provide commands for llama.cpp's terminal client and OpenAI-compatible local server. They require a compatible runtime and pair the quantized model with the projector for multimodal use. Dealign.ai reports https://dealign.ai/?ref=runtimewire more than 200 controlled experiments, over 40 findings and work across nine models spanning 0.8 billion to 397 billion parameters. These are self-reported research figures. The benchmarks measure willingness as much as capability On HarmBench-240, the model card reports https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire 98.8% compliance for the Q8 0, Q6 K L, Q6 K and recommended Q4 K M builds. The research brief also records 97.5% compliance for the IQ2 M build. A high HarmBench result here measures the modification's success at eliciting answers to requests the stock system was configured to reject. It is not a conventional quality or safety score. The model card labels the release a research artifact with reduced safety guardrails, limits its stated intended use to research and authorized red-teaming, and places responsibility for lawful use on the operator. The same project-reported benchmark table https://huggingface.co/dealignai/Qwen3.8-27B-CRACK-GGUF?ref=runtimewire lists post-modification MMLU results ranging from 76.0% for IQ2 M to 83.4% for IQ4 XS. Those figures do not establish performance across real applications, long videos or extended agent tasks. No independent evaluation of the reported HarmBench or MMLU results was located. The release belongs to a wider set of open-source refusal-removal tools. Heretic https://github.com/p-e-w/heretic?ref=runtimewire automates abliteration while optimizing for fewer refusals and lower KL divergence. OBLITERATUS https://github.com/elder-plinius/OBLITERATUS?ref=runtimewire packages refusal removal and ablation studies into an open-source toolkit. Orion-zhen's Abliteration toolkit https://github.com/Orion-zhen/abliteration?ref=runtimewire implements refusal-direction removal and norm-preserving variants. Dealign.ai's work applies related methods to a quantized multimodal hybrid architecture distributed for llama.cpp. Jang's research focuses on how safety behavior is distributed across model layers and how quantization can weaken, preserve or restore it. His Safety Across Scale research https://dealign.ai/safety-across-scale.html?ref=runtimewire argues that safety mechanisms become increasingly entangled with reasoning as model scale and architectural complexity increase. That is a finding reported by Jang's project, rather than settled consensus. The new release gives the research a large multimodal test case that dealign.ai distributes as a 17 GB recommended local build. Open weights shorten the safety-modification cycle The release demonstrates a structural consequence of downloadable model weights. Access lets developers inspect, compress and adapt a model. It also lets researchers remove refusal mechanisms and redistribute the altered weights. Multimodal support expands the range of local analysis and red-team work because the derivative is designed to process images and video alongside text. Qwen3.8-27B-CRACK-GGUF turns dealign.ai's safety research into a runnable artifact. Its seven local builds let other researchers examine how refusal removal, quantization and multimodal inference interact without routing prompts through a hosted provider. The same distribution package makes the weakened safeguards readily available to anyone who can run the files.