A unified version of Qwen3.5-9B has been released, obtained by combining Jackrong's Claude 4.6 Opus-based model with an uncensored version developed by HauhauCS.

Key Features

  • Integration: Combines the reasoning capabilities of Qwen3.5-9B with Claude 4.6 Opus training data.
  • Content Filter: Uses a less restrictive content filter than standard models.
  • Performance: Optimized for local execution, with a focus on efficiency on less powerful hardware. On an RTX 3060, 42 tokens per second are achieved in LM Studio.
  • Format: Available in GGUF format, suitable for inference on CPUs and GPUs with limited memory.

The model is designed for those who want to run large language models locally, without the typical restrictions of more conservative models.