Qwen3.5-9B-Claude-4.6-HighIQ-THINKING-HERETIC-UNCENSORED
What is it?
A fine‑tuned version of Qwen 3.5 9B that thinks like Claude 4.6 instead of the original Qwen.
It’s also fully uncensored – no refusals, no restrictions.
Key improvements
- Better reasoning – the model’s internal “thinking” is now clearer and more useful.
- Uncensored – it follows instructions freely (tested: only 6 refusals out of 100 vs. 100/100 for the original).
- Vision works – you can give it images (and it handles them well).
- Original performance is preserved – benchmarks stayed almost identical.
Quick specs
| Feature | Value |
|---|---|
| Parameters | 9 billion |
| Context length | 256k tokens (default) |
| Quantization | Use at least q4ks or IQ3S |
| Repetition penalty | Keep it at 1.0 (off) |
| Modes | Thinking (default) or Instruct (no‑thinking) |
How to use it
- It works with standard tools (SGLang, vLLM, Transformers).
- For thinking mode (default): set
temperature=1.0, top_p=0.95, top_k=20. - For instruct mode (no thinking): add
"chat_template_kwargs": {"enable_thinking": False}.
Simple example (Python)
from openai import OpenAI
client = OpenAI(base_url="http://localhost:8000/v1", api_key="EMPTY")
response = client.chat.completions.create(
model="Qwen/Qwen3.5-9B",
messages=[{"role": "user", "content": "Explain quantum computing simply"}],
temperature=1.0,
top_p=0.95,
extra_body={"top_k": 20}
)
print(response.choices[0].message.content)
Why choose this model?
- You want intelligent, uncensored responses.
- You like Claude‑style reasoning but in a smaller, open‑source package.
- You need vision + text in one model.