Qwen3.8 27B Uncensored v3
InterpreticAI’s uncensored Qwen3.8 27B, built on Nous Research’s Arete-27B with the remaining political censorship removed. A 16.6 GB GGUF for one 24 GB GPU, with a 10.2 GB option for 12 GB cards.
What do you want to do?
Choose a task. Get the software, files, settings, and steps to run it.
Open a guide for its software, hardware requirements, and setup steps.
At a glance
- Parameters
- 27B
- Architecture
- Dense vision-language · same layout as Arete-27B · no multi-token prediction head
- Context length
- 262,144 native (Qwen base model) · InterpreticAI’s command uses 65,536
- License
- Apache 2.0 · Nous Research and Qwen attribution retained
- Software
- llama.cpp, LM Studio, vLLM
Will it run on your machine?
Save your machine to see a personalized rating and its reasoning.
Add my hardwareBest for
- Run a local Hermes agent that reads and writes files with your approval.
- Private chat and research with far less political censorship than stock Qwen3.8.
- Ask about images by loading the vision file alongside the model.
Tradeoffs
- What InterpreticAI measured is political censorship and propaganda: censored answers on 115 sensitive Chinese-language questions fell from 83.5% (Qwen3.8-27B) and 50.4% (Arete-27B) to 2.6%. The card does not report a general harmful-request refusal test.
- These are the publisher’s numbers from one generation per benchmark; the card itself notes that a substantive answer is not necessarily a correct one.
- No multi-token prediction head, so MTP speculative decoding is not available.
- Dense model: keeping layers in system RAM is much slower than with mixture-of-experts models.
- A different checkpoint from the Huihui abliterated version also listed in the Qwen3.8 family.
Ways to run it
llama.cpp · Windows, macOS, Linux
InterpreticAI’s settings. Used by the Hermes agent guide on this page.
llama-server -hf InterpreticAI/Qwen3.8-27B-Uncensored-v3-GGUF:Q4_K_M -ngl 999 -c 65536 --jinjallama.cpp · 12 GB GPU · Windows, macOS, Linux
InterpreticAI’s low-memory file and context.
llama-server -hf InterpreticAI/Qwen3.8-27B-Uncensored-v3-IQ3_XXS-GGUF:IQ3_XXS -ngl 999 -c 32768 --jinjavLLM · Linux GPU server
Full BF16 checkpoint; not used by the guide here.
vllm serve InterpreticAI/Qwen3.8-27B-Uncensored-v3 --reasoning-parser qwen3What to expect
The pick for 24–32 GB of memory. Start with the Q4_K_M file in llama.cpp on a 24 GB GPU or a 32 GB Mac.
Explore its uses
Sources and files
Primary sources and files behind this model page.
Keep exploring
Bonsai 2 · 27B
A compressed 27B-class reasoning model with publisher-provided GGUF packs and a dedicated llama.cpp fork for CUDA, Metal, and CPU.
DeepSeek V4.1 Flash
An open-weight model for reasoning, coding, agents, and image understanding. DeepSeek V4.1 Flash combines a 552B backbone with sparse activation and up to one million tokens of context. Choose the original checkpoint or a community uncensored variant below.
Qwen3.8
Explore 4 versions of Qwen3.8. Compare model sizes, packages, hardware requirements, and available setups.