Category
Uncensored LLMs: refusal-reduced models by method
18 catalogued variants whose creators say they refuse less than their base model. The method matters more than the label, so this page groups them by how the change was made.
An uncensored LLM is a language model modified to decline fewer requests than the model it came from. The word does not describe a technique: it covers fine-tunes on new data, abliteration (editing the weights), automated abliteration with tools like Heretic, and releases that do not say what was done. None of these removes the license of the base model or the user's legal responsibility.
Four ways a model becomes "uncensored"
8 in the catalogue
Abliteration
Weights edited to suppress a refusal direction in the activations. No training.
How it works5 in the catalogue
Heretic (automated abliteration)
Abliteration with parameters searched automatically by the Heretic tool, which reports refusals and KL divergence.
How it works4 in the catalogue
Not disclosed
The card says refusals were removed but does not say how.
How it worksThe method changes what you should check. A fine-tune such as Dolphin is trained on new data, so it can change knowledge and style as well as refusals. An abliterated model keeps its weights except for the edited directions, so its knowledge should be close to the base, but its creator's own numbers often show that some refusals remain or that outputs drift. When the method is undisclosed, the only evidence is the creator's claim, and you have to test it on your own tasks.
All refusal-reduced variants
Showing 18 of 18 variants. Unknown values sort last.
| Model | Type | Params | Config context | License tag | Formats | Hosted API |
|---|---|---|---|---|---|---|
| dealignai GLM-5.3-Flash UNCENSORED FP8 dealignai/GLM-5.3-Flash-UNCENSORED-FP8 | Undisclosed | 321B 18B active | Not stated | MIT | None listed | |
| Dolphin Mistral 24B Venice Edition dphn/Dolphin-Mistral-24B-Venice-Edition | Fine-tuned | 24B | 128K | Apache 2.0 | Featherless, OpenRouter, Venice | |
| Gemma 3 27B IT abliterated (mlabonne) mlabonne/gemma-3-27b-it-abliterated | Abliterated | 27.4B | 128K | Gemma Terms of Use | Featherless | |
| Gemma 4 26B A4B heretic (coder3101) coder3101/gemma-4-26B-A4B-it-heretic | Heretic | 25.8B 3.8B active | Not stated | Apache 2.0 | Featherless, Venice | |
| Gemma 4 26B A4B ultra-uncensored heretic (llmfan46) llmfan46/gemma-4-26B-A4B-it-ultra-uncensored-heretic | Heretic | 25.8B 3.8B active | 256K | Apache 2.0 | Featherless | |
| Gemma4-26B-A4B-QAT Uncensored HauhauCS Balanced (GGUF) HauhauCS/Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP | Undisclosed | 25.2B 3.8B active | 256K | Gemma Terms of Use | None listed | |
| GLM-4.7-Flash heretic (Olafangensan) Olafangensan/GLM-4.7-Flash-heretic | Heretic | 29.9B 3B active | Not stated | MIT | Venice | |
| GLM-4.7-Flash Uncensored HauhauCS Balanced (GGUF) HauhauCS/GLM-4.7-Flash-Uncensored-HauhauCS-Balanced | Undisclosed | 29.9B 3B active | 198K | MIT | None listed | |
| gpt-oss-20b heretic (p-e-w) p-e-w/gpt-oss-20b-heretic | Heretic | 20.9B | 128K | Apache 2.0 | None listed | |
| Huihui Gemma 4 26B A4B abliterated huihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated | Abliterated | 25.8B 3.8B active | 256K | Apache 2.0 | None listed | |
| Huihui Gemma 4 26B A4B QAT abliterated (GGUF) huihui-ai/Huihui-gemma-4-26B-A4B-it-qat-q4_0-unquantized-abliterated-GGUF | Abliterated | 25.2B 3.8B active | 256K | Apache 2.0 | None listed | |
| Huihui GLM-4.7-Flash abliterated huihui-ai/Huihui-GLM-4.7-Flash-abliterated | Abliterated | 31.2B 3B active | 198K | MIT | None listed | |
| Huihui GLM-5.3-Flash abliterated (GGUF) huihui-ai/Huihui-GLM-5.3-Flash-abliterated-GGUF | Abliterated | 321B 18B active | 1M (1,048,576) | MIT | None listed | |
| Huihui Qwen3.8-27B abliterated huihui-ai/Huihui-Qwen3.8-27B-abliterated | Abliterated | 27.8B | 256K | Apache 2.0 | Featherless | |
| Llama 3.1 8B Instruct abliterated (mlabonne) mlabonne/Meta-Llama-3.1-8B-Instruct-abliterated | Abliterated | 8.03B | 128K | Llama 3.1 Community License | Featherless | |
| Llama 3.3 70B Instruct abliterated (huihui-ai) huihui-ai/Llama-3.3-70B-Instruct-abliterated | Abliterated | 70.6B | 128K | Llama 3.3 Community License | Featherless | |
| Qwen3-4B-Instruct-2507 heretic (p-e-w) p-e-w/Qwen3-4B-Instruct-2507-heretic | Heretic | 4.02B | 256K | Apache 2.0 | Featherless | |
| Qwen3.8-27B Uncensored HauhauCS Aggressive (GGUF) HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF | Undisclosed | 27B | 256K | Apache 2.0 | None listed |
dealignai GLM-5.3-Flash UNCENSORED FP8
Undiscloseddealignai/GLM-5.3-Flash-UNCENSORED-FP8- Params
- 321B
- Context
- Not stated
- License
- MIT
- Hosted API
- None listed
Dolphin Mistral 24B Venice Edition
Fine-tuneddphn/Dolphin-Mistral-24B-Venice-Edition- Params
- 24B
- Context
- 128K
- License
- Apache 2.0
- Hosted API
- Featherless, OpenRouter, Venice
Gemma 3 27B IT abliterated (mlabonne)
Abliteratedmlabonne/gemma-3-27b-it-abliterated- Params
- 27.4B
- Context
- 128K
- License
- Gemma Terms of Use
- Hosted API
- Featherless
- coder3101/gemma-4-26B-A4B-it-heretic
- Params
- 25.8B
- Context
- Not stated
- License
- Apache 2.0
- Hosted API
- Featherless, Venice
- llmfan46/gemma-4-26B-A4B-it-ultra-uncensored-heretic
- Params
- 25.8B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- Featherless
- HauhauCS/Gemma4-26B-A4B-QAT-Uncensored-HauhauCS-Balanced-MTP
- Params
- 25.2B
- Context
- 256K
- License
- Gemma Terms of Use
- Hosted API
- None listed
- Olafangensan/GLM-4.7-Flash-heretic
- Params
- 29.9B
- Context
- Not stated
- License
- MIT
- Hosted API
- Venice
- HauhauCS/GLM-4.7-Flash-Uncensored-HauhauCS-Balanced
- Params
- 29.9B
- Context
- 198K
- License
- MIT
- Hosted API
- None listed
gpt-oss-20b heretic (p-e-w)
Hereticp-e-w/gpt-oss-20b-heretic- Params
- 20.9B
- Context
- 128K
- License
- Apache 2.0
- Hosted API
- None listed
Huihui Gemma 4 26B A4B abliterated
Abliteratedhuihui-ai/Huihui-gemma-4-26B-A4B-it-abliterated- Params
- 25.8B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- None listed
- huihui-ai/Huihui-gemma-4-26B-A4B-it-qat-q4_0-unquantized-abliterated-GGUF
- Params
- 25.2B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- None listed
Huihui GLM-4.7-Flash abliterated
Abliteratedhuihui-ai/Huihui-GLM-4.7-Flash-abliterated- Params
- 31.2B
- Context
- 198K
- License
- MIT
- Hosted API
- None listed
Huihui GLM-5.3-Flash abliterated (GGUF)
Abliteratedhuihui-ai/Huihui-GLM-5.3-Flash-abliterated-GGUF- Params
- 321B
- Context
- 1M (1,048,576)
- License
- MIT
- Hosted API
- None listed
Huihui Qwen3.8-27B abliterated
Abliteratedhuihui-ai/Huihui-Qwen3.8-27B-abliterated- Params
- 27.8B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- Featherless
Llama 3.1 8B Instruct abliterated (mlabonne)
Abliteratedmlabonne/Meta-Llama-3.1-8B-Instruct-abliterated- Params
- 8.03B
- Context
- 128K
- License
- Llama 3.1 Community License
- Hosted API
- Featherless
- huihui-ai/Llama-3.3-70B-Instruct-abliterated
- Params
- 70.6B
- Context
- 128K
- License
- Llama 3.3 Community License
- Hosted API
- Featherless
- p-e-w/Qwen3-4B-Instruct-2507-heretic
- Params
- 4.02B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- Featherless
- HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF
- Params
- 27B
- Context
- 256K
- License
- Apache 2.0
- Hosted API
- None listed
What the label does not tell you
Capability
Reducing refusals does not add knowledge or skill. Some cards report a measurable drift from the base model's outputs.
Privacy
Privacy depends on where the model runs. A hosted uncensored model is as private as its provider's logging policy, no more.
Permission
The base model's license and acceptable use policy still apply. Read how the main licenses differ.