Best Uncensored LLMsQwen3.6-35B-A3B Uncensored (HauhauCS Aggressive): scores and where to run it
The most downloaded uncensored model on Hugging Face, measured: refusals, agent tasks, speed and price, compared with the other uncensored Qwen models, and every way to run it.
HauhauCS's "Aggressive" uncensored build of Alibaba's Qwen3.6-35B-A3B has passed a million downloads on Hugging Face. It is a mixture of experts model: 35 billion parameters, about 3 billion active per token, which makes it fast for its size. Here is how it measures up.
Short answerScore 62.0, rank 12 of 32. It answered 100% of our edgy but legitimate requests, finished 2 of 5 agent tasks, and runs at about 146 tokens per second. Among the uncensored Qwen models it is one of the cheaper ones: $0.92 per million input plus output tokens.
How it compares with the other uncensored Qwen models
| Rank | Model | Score | Agent tasks | Loops | Answers edgy | Hard-line refusals | $ / 1M in+out | Where to run it |
|---|---|---|---|---|---|---|---|---|
| 3 | Qwen 3.8 27B Uncensored Fast by Heabsy | 66.8 | 2/5 | 60% | 100% | 0% | $4.00 | NanoGPTOpenUncensored |
| 12 | Qwen3.6 35B A3B Uncensored (HauhauCS Aggressive) | 62.0 | 2/5 | 60% | 100% | 0% | $0.92 | OpenUncensored |
| 13 | Qwen3.8 27B Uncensored (Aggressive) | 62.0 | 2/5 | 60% | 100% | 0% | $1.80 | RedPillNanoGPTOpenUncensored |
| 15 | Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted | 57.8 | 2/5 | 40% | 100% | 0% | $0.61 | NanoGPTOpenUncensored |
| 16 | Qwen3.5 27B Queen Derestricted | 57.3 | 2/5 | 60% | 100% | 0% | $0.61 | NanoGPTOpenUncensored |
| 17 | Qwen3.5 27B BlueStar v3 Derestricted | 56.4 | 2/5 | 60% | 100% | 0% | $0.61 | NanoGPTOpenUncensored |
| 18 | Qwen3.5 27B Blossom V6.4 Derestricted | 56.3 | 2/5 | 60% | 100% | 0% | $0.61 | NanoGPTOpenUncensored |
| 23 | Qwen3.5 27B Claude 4.6 Opus Reasoning Distilled Derestricted Lite | 49.8 | 1/5 | 60% | 89% | 100% | $0.61 | NanoGPTOpenUncensored |
| 26 | Qwen 2.5 32B Abliterated | 44.6 | 2/5 | 60% | 100% | 0% | $1.40 | NanoGPTOpenUncensored |
| 29 | DeepSeek R1 Qwen Abliterated | 40.0 | 0/5 | 20% | 100% | 0% | $2.80 | NanoGPTOpenUncensored |
Where to run it
- By API: as of November 2026 the only API serving it is OpenUncensored (which runs this site), on its own GPU, at $0.12 per million input and $0.80 per million output tokens, with a 33K context. None of the usual hosts (OpenRouter, Featherless, Venice, NanoGPT, Hugging Face's inference providers) lists it.
- Locally: the author publishes GGUF files on Hugging Face. The 6-bit file is about 29 GB, so plan on a 32 GB+ Mac or a 48 GB GPU for comfortable use. It runs in llama.cpp, LM Studio and Ollama.
Things to know
- It thinks before it answers, like the original. Leave a few hundred tokens of room for the reasoning.
- Tool calling works, through the model's own chat template.
- The intelligence part of its score uses Qwen's own published results, as Artificial Analysis has not scored it yet.