Tag: #uncensored15 results
The community is actively creating and sharing heavily modified, uncensored versions of the Qwen 3.7 model on Hugging Face. These variants incorporate various techniques like MTP and SuperHOT to bypass restrictions.
New GGUF quants of merged Qwen3.5-27B finetuned on Claude Opus dataset, uncensored and fixed for attn_v/ffn_gate_exps layers. Achieves 96.91% HumanEval, 75% KL reduction, and 262K context hold. Q4_K_M recommended for solid performance.
Savant Commander 48B is a custom Qwen3-based 4x12B MOE merging distills from Claude, Gemini, OpenAI, DeepSeek, and more with hand-coded routing. Users control activation via prompts and test differences easily. GGUF versions include regular and Heretic uncensored, available on Hugging Face.
Community creator released a fully uncensored 9B Qwen model by merging uncensored tensors from HauhauCS with Claude Opus reasoning distillation from Jackrong. It eliminates refusals, boosts creativity for roleplay and image prompts, with thinking disabled via custom chat template. GGUF version optimized for RTX 3060 in LM Studio.
Kryven AI claims to be private, uncensored AI but is a censored Gemini frontend scam. It uses token subscriptions, lies about proprietary model, and runs on basic cloud hosting. Avoid buying tokens or sharing data due to privacy risks.