Uncensored Multimodal Power: Qwen3.6-35B-A3B Aggressive GGUF Model Takes the Lead

The Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive model is a 35‑billion‑parameter Mixture‑of‑Experts (MoE) transformer that retains the full capabilities of the original Qwen3.6‑35B‑A3B while removing all safety refusals. Tagged as "image-text-to-text," it natively handles multimodal inputs—text, images, and video—across English, Chinese, and other languages. The repository offers a suite of custom K_P quantizations (Q8_K_P down to Q2_K_P) and an accompanying mmproj file, all packaged in the GGUF format for seamless use with llama.cpp, LM Studio, Jan, koboldcpp, and other compatible runtimes.

Key technical specs include 256 experts with 8 routed per token, a hybrid attention architecture (linear to full softmax 3:1), 40 layers, and a massive 262K native context window. The "Aggressive" variant is explicitly marketed as fully unlocked, generating content without refusals and only occasional short, baked‑in disclaimers. The README emphasizes the importance of keeping at least 128K context for optimal "thinking" mode and provides recommended temperature and top‑p settings for both general and coding tasks. The model’s licensing under Apache‑2.0 makes it free for commercial and research use.

Because the model is multilingual and uncensored, it is especially attractive for developers needing candid, unrestricted multimodal generation—whether for creative applications, detailed image analysis, or bilingual content creation. Its high‑quality K_P quantizations promise near‑lossless performance while keeping file sizes manageable for local deployment.