@IndieDevHailey: Guys, local players can now experiment without boundaries. Alibaba's Qwen3.8-27B has been completely obliterated (safety restrictions removed), with the V2 version achieving zero rejection. Even better, MMLU (Massive Multitask Language Understanding) score jumped from 85.3% to 86.3%.
Summary
Alibaba's Qwen3.8-27B model's safety restrictions have been completely removed, with the V2 version achieving zero rejection, and the MMLU score improved to 86.3%, using complementary ablation blending to achieve uncensored and stronger performance.
View Cached Full Text
Cached at: 08/21/26, 09:09 AM
Folks, local enthusiasts can now play without boundaries.
Alibaba’s Qwen3.8-27B has been completely OBLITERATED—safety restrictions fully removed, with the V2 version having zero refusals.
Better yet, the MMLU (Massive Multitask Language Understanding, a benchmark measuring a model’s knowledge and reasoning capabilities) score rose from 85.3% to 86.3%, surpassing the original version.
By using Complementary Abliteration Blending—a technique that combines two different ablation methods to offset their respective weaknesses—safety alignment was stripped clean, yet performance actually improved.
Truly uncensored, and still highly capable.
Similar Articles
@shangdu2005: 比上个版本更加无敌的。 Qwen3.8 27B越狱无审查版本。 这两个自行测试，谨慎使用。 1.orcarouter https://huggingface.co/orcarouter/Qwen3.8-27B-Uncensored-FP8…
This article presents an uncensored version of the Qwen3.8 27B AI model, which has been modified to remove safety refusals and quantized to FP8 for efficiency, intended for research purposes.
@Lonely__MH: Come on! Do as you please 2.0 version is here! Qwen 3.8-27B cracked version direct API call! No local deployment needed! Completely free your computer! Yesterday, I shared a local tutorial for Qwen-3.8 27B uncensored version, and many followers found it too high of a barrier. So, I'm bringing the API right to you…
This article announces the cracked version of Qwen 3.8-27B, offering direct API call functionality without local deployment, aiming to simplify access to AI models.
LOL, my timeline was flooded with this guy's shock. Someone just bypassed the safety guardrails of Qwen3.8-27B. OrcaRouter had previously released a Qwen3.8-27B-Uncensored-FP8 on Hugging Face. They used Abliteration…
By using Abliteration to modify the weights, someone published an uncensored version of Qwen3.8-27B, drastically lowering the refusal rate and preserving model performance, which has sparked debates on AI safety.
@GoSailGlobal: Today, August 3, Alibaba officially released Qwen3.8. With 2.4 trillion parameters and 1M context, coding and office capabilities have improved dramatically in this version, pushing it into the global top tier overall. The preview version ran for two weeks, and today it officially graduated. I immediately put it through a hardcore real-world test—let me lay out the results and process. I came up with a pretty tough challenge: single-file HTML hand...
Alibaba officially released Qwen3.8, with 2.4 trillion parameters and 1M context. Coding and office capabilities have been significantly improved, entering the global top tier. The author conducted a hardcore real-world test using a single-file HTML N-body simulation.
@sanbuphy: K2.6 successfully downloaded and deployed the Qwen3.5-0.8B model locally on a Mac, using the niche Zig language to implement and optimize inference, demonstrating the new model’s generalization ability. After 4,000+ tool calls and 12+ hours of continuous operation, K2.6 iterated 14 times…
K2.6 successfully downloaded and deployed the Qwen3.5-0.8B model locally on a Mac, using the niche Zig language to implement and optimize inference, demonstrating the new model’s generalization ability. After 4,000+ tool calls and 12+ hours of continuous operation, K2.6 iterated 14 times, boosting throughput from ~15 tokens/s to ~193 tokens/s, ultimately achieving 20% faster inference than LM Studio.