An Ollama uncensored model is a GGUF you pull and run on your own GPU: llama2-uncensored, the Dolphin tags, wizardlm-uncensored and the abliterated DeepSeek distills. This guide lists every one of them from the Ollama library as read on 27 October 2026, with its size, and puts the same or newer build on this key next to it, with the live price per million tokens and what 100 messages cost. The last section is the two-line change that points Open WebUI or AnythingLLM at the key instead of Ollama.
If the words abliterated and uncensored are new, abliterated vs uncensored explains the difference in one page.
Every ollama uncensored model on the library today
The Ollama library's search for "uncensored" on 27 October 2026 listed eight official tags and a dozen community ones. The official tags are the rows below. Each one links to its Ollama page; the size column is what that page prints. Every one of them was last updated one to two years ago.
| Ollama tag | Size class | What the page says |
|---|---|---|
llama2-uncensored | 7B (3.8 GB), 70B (39 GB) | Llama 2, by George Sung and Jarrad Hope; 2.7M pulls, updated 2 years ago |
dolphin-llama3 | 8B (4.7 GB), 70B (40 GB), 8K context | Dolphin 2.9 on Llama 3, by Eric Hartford; 2.2M pulls, updated 2 years ago |
dolphin-mistral | 7B (4.1 GB), 32K context | Dolphin 2.8 on Mistral 7B; 1.8M pulls, updated 2 years ago |
dolphin3 | 8B (4.9 GB), 128K context | Dolphin 3.0 on Llama 3.1 8B; 4.2M pulls, updated 1 year ago |
wizardlm-uncensored | 13B (7.4 GB), 4K context | 666.5K pulls, updated 2 years ago |
huihui_ai/deepseek-r1-abliterated | 1.5B to 70B (1.1 GB to 43 GB), 128K context | DeepSeek R1 distills, abliterated by huihui_ai; 748.7K pulls, updated 1 year ago |
Three more official tags came up in the same search and are not in the table because this key has no counterpart for them: dolphin-mixtral (8x7B and 8x22B), wizard-vicuna-uncensored (7B to 30B) and everythinglm (13B, 16K context). The community tags were mostly abliterated Qwen 3.8 and Gemma 4 builds under personal namespaces, such as eriktonon/qwen3.8-uncensored at 9B and 27B.
ollama run llama2-uncensored# or any tag in the table: ollama run huihui_ai/deepseek-r1-abliterated:14bThe same builds by API, with no GPU
Each Ollama tag has a build on this key that is the same family or a newer one, served by API and priced per token. The 70B Llama 2 tag needs 39 GB of weights; llama-3.3-70b-abliterated is Llama 3.3 at the same size with the refusal direction removed. The two Dolphin tags were 7B and 8B; dolphin-mistral-24b-venice is Dolphin on Mistral Small 24B. The DeepSeek R1 distills get deepseek-v4.1-flash-abliterated, a current DeepSeek with no retraining.
| Ollama tag | Our id | Input / M | Output / M | 100 messages |
|---|---|---|---|---|
llama2-uncensored | llama-3.3-70b-abliterated | $2.80 | $2.80 | $0.64 |
dolphin-llama3 | llama-3.3-70b-abliterated | $2.80 | $2.80 | $0.64 |
dolphin-mistral | dolphin-mistral-24b-venice | $0.50 | $2.25 | $0.17 |
dolphin3 | dolphin-mistral-24b-venice | $0.50 | $2.25 | $0.17 |
wizardlm-uncensored | qwen3.8-27b-uncensored | $1.25 | $3.75 | $0.36 |
huihui_ai/deepseek-r1-abliterated | deepseek-v4.1-flash-abliterated | $0.8488 | $1.6975 | $0.22 |
The daily check column below comes from the daily real-model check, which sends a fixed prompt to every model each day and records whether the answer still matches the model on the label. An Ollama pull has no equivalent; the GGUF on your disk is what it is.
| Wire formats | |||||
|---|---|---|---|---|---|
| Llama 3.3 70B Abliterated | Abliterated | $2.80 | $2.80 | 33K | |
| Dolphin Mistral 24B Venice Edition | Uncensored | $0.50 | $2.25 | 33K | |
| Qwen3.8 27B Uncensored | Uncensored | $1.25 | $3.75 | 33K | |
| DeepSeek V4.1 Flash Abliterated | Abliterated | $0.8488 | $1.6975 |
| Model id | Daily check | Context window |
|---|---|---|
llama-3.3-70b-abliterated | Steady since Oct 22 | 32,768 tokens |
dolphin-mistral-24b-venice | Steady since Oct 20 | 32,768 tokens |
qwen3.8-27b-uncensored | Steady since Oct 22 | 32,768 tokens |
deepseek-v4.1-flash-abliterated | Steady since Oct 19 | not published |
The Llama rows sit on the uncensored Llama page, the DeepSeek row on the uncensored DeepSeek page, and the Qwen row on the uncensored Qwen page, each with the rest of its family.
Ollama deepseek abliterated: the local tag and the API build
The tag people mean by "ollama deepseek abliterated" is huihui_ai/deepseek-r1-abliterated. Its page on 27 October 2026 listed seven sizes from 1.5B to 70B, all with a 128K context, 748.7K pulls and a last update one year ago. The 32B quant is 20 GB and the 70B is 43 GB, so the sizes most laptops run are the 7B, 8B and 14B distills.
On this key the DeepSeek line is deepseek-v4.1-flash-abliterated, the current V4.1 Flash with the refusal direction removed, and deepseek-v4.1-flash-abliterated-fast, the same build on a faster lane. Neither is a distill. The request is one host change away from Ollama's own OpenAI-compatible endpoint, which its docs put at http://localhost:11434/v1.
# Ollama's own OpenAI-compatible endpoint, from its docscurl http://localhost:11434/v1/chat/completions \ -H "Content-Type: application/json" \ -d '{"model": "dolphin-llama3:8b", "messages": [{"role": "user", "content": "Hello"}]}'# The same request against this key: change the host and the model idcurl https://api.uncensoredllm.ai/v1/chat/completions \ -H "Authorization: Bearer sk-ullm-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "llama-3.3-70b-abliterated", "messages": [{"role": "user", "content": "Hello"}]}'The second request is billed per token at the row above and needs no download, no VRAM and no quant choice. The key starts with sk-ullm-, is created in the console and is shown once.
Dolphin llama 3 and llama2-uncensored: what to run instead
dolphin-llama3 is Dolphin 2.9 on Llama 3, 8B and 70B, with an 8K context. llama2-uncensored is older still: Llama 2 at 7B and 70B, 4K context. Both pages were last updated two years ago. They still run, and for a 7B on a laptop they are fine.
For a 70B the API build is the practical route. The 70B GGUF is 39 to 40 GB on disk before it loads. llama-3.3-70b-abliterated is Llama 3.3 70B Instruct with huihui-ai's abliteration, the window in the table above, and a price per token on the same row. A 100-message evening on it is the figure in the last column; no electricity, no download.
For a Dolphin, dolphin-mistral-24b-venice is Cognitive Computations' Dolphin on Mistral Small 24B, three times the 7B tag's size; its window is in the table above.
Point Open WebUI or AnythingLLM at this key instead of Ollama
Both apps talk to Ollama through an OpenAI-compatible connection, and both take a second connection of the same kind. Two lines change: the URL and the key. The model picker then fills from /v1/models with every uncensored id the key can reach.
Open WebUI
Under Settings > Admin > Connections, add an OpenAI API connection, per its connection doc. Leave the Ollama connection on if you want both in one picker.
URL https://api.uncensoredllm.ai/v1API Key sk-ullm-live-…The long version, with a short picker set on the key and one preset per build, is the Open WebUI guide.
AnythingLLM
In Settings, switch the LLM provider from Ollama to the provider AnythingLLM's docs call OpenAI (generic), the developer-facing one, and fill its base URL, key and model name:
Base URL https://api.uncensoredllm.ai/v1API key sk-ullm-live-…Model llama-3.3-70b-abliterated- Base URL
- https://api.uncensoredllm.ai/v1Ends in /v1, no trailing slash
- Key
- sk-ullm-live-…Sent as a Bearer token
- Model
- llama-3.3-70b-abliteratedAny id from /v1/models
Check the key from a terminal first; the call is free and lists exactly the ids the picker will show.
curl https://api.uncensoredllm.ai/v1/models -H "Authorization: Bearer sk-ullm-live-…"SillyTavern, Janitor AI and the other roleplay front-ends take the same two lines; the docs page has each one's fields.
Questions
Because the model Ollama is serving is a stock instruct build. llama3, mistral, qwen and gemma on the library ship with their alignment intact, and AnythingLLM sends your prompt to whichever tag is selected. Pull one of the uncensored tags above and select it, or switch the provider to this API and pick an abliterated id; the refusals are removed from the weights, so no system prompt is needed. See AI with no restrictions.
For a laptop, the 7B and 8B tags in the first table are the ones that fit: dolphin3 has the widest context at 128K. For a 70B, the download is 39 to 43 GB; the API rows above run the same size class per token with no hardware. The ranked list across every build on this key is the leaderboard. See the uncensored LLM leaderboard.
The Ollama search on 27 October 2026 showed community abliterated Qwen 3.8 tags such as eriktonon/qwen3.8-uncensored at 9B and 27B, none of them official. On this key qwen3.8-27b-uncensored is the general build and the Qwen roleplay tunes are on the uncensored Qwen page. Adults only, fiction only, per the acceptable use page. See the uncensored Qwen page.
Yes. The tag was on the library on 27 October 2026 with 2.7M pulls, 7B and 70B sizes and a last update two years ago. It is Llama 2, so its context is 4K; llama-3.3-70b-abliterated is the current Llama at the same size on this key.
It is the R1 distills: Qwen and Llama models trained on R1 outputs, abliterated by huihui_ai, from 1.5B to 70B. The 671B DeepSeek weights do not fit a consumer card. deepseek-v4.1-flash-abliterated on this key is a DeepSeek build, not a distill, and the daily check records whether it still answers as one. See the daily real-model check.
Yes. Open WebUI takes several connections at once. Keep the Ollama connection for the local tags and add https://api.uncensoredllm.ai/v1 with your key as an OpenAI API connection; both sets of ids appear in one picker.