Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive

Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive is a vision-language model (VLM) that performs text, image, and video understanding tasks with 0 out of 465 test prompts eliciting refusals. It is an uncensored, aggressive variant of Qwen3.6-35B-A3B, designed to fully unlock model capabilities without safety guardrails.

Model Architecture

• Parameters: 35B total, ~3B active per forward pass (Mixture-of-Experts)
• Experts: 256 experts, 8 routed per token
• Attention: Hybrid linear attention + full softmax attention (3:1 ratio)
• Layers: 40
• Context length: 262K tokens natively
• Base model: Qwen/Qwen3.6-35B-A3B

Available Quantizations

The model is offered in multiple GGUF quantizations generated with importance matrix for optimal quality preservation on abliterated weights. A companion mmproj file (mmproj-Qwen3.6-35B-A3B-Uncensored-HauhauCS-Aggressive-f16.gguf) is required for vision support.

Recommended Settings

Based on the original Qwen authors’ guidance:

• Thinking mode (default): temperature=1.0, top_p=0.95, top_k=20, min_p=0, presence_penalty=1.5; for coding/precise tasks, temperature=0.6, presence_penalty=0.
• Non-thinking mode: temperature=0.7, top_p=0.8, top_k=20, presence_penalty=1.5 (general); temperature=1.0, top_p=1.0, top_k=40, presence_penalty=2.0 (reasoning).
• Keep at least 128K context to preserve thinking capabilities.

This model is hosted by Gigarouter as an OpenAI-compatible API. No local installation is required.