OpenRouter vs OpenUncensored
Uncensored models on OpenRouter, and where the rest are.
OpenRouter is the biggest model API, but uncensored models are a small corner of it. Here is what it has, what it does not, and how the two compare. OpenRouter's list checked November 10, 2026.
Short answer
Of OpenRouter's 469 models, one is explicitly uncensored and about ten are roleplay fine-tunes. The abliterated and heretic versions of DeepSeek V4.1 Flash, GLM 5.3, Qwen 3.8 27B, Qwen3.6 35B A3B and similar models are not there. OpenUncensored serves 44 uncensored and abliterated text models, plus all of OpenRouter's roleplay models, on one OpenAI compatible key, and scores 32 of them on refusals and agent tasks.Side by side
OpenRouter and OpenUncensored.
| OpenRouter | OpenUncensored | |
|---|---|---|
| Models in total | 469 | 77 |
| Explicitly uncensored or abliterated text models | 1 | 44 |
| Roleplay fine-tunes (Hermes, Magnum, Euryale, ...) | about 10 | the same ones |
| Abliterated DeepSeek, GLM, Qwen, Gemma, MiMo | no | yes |
| Uncensored image models | none listed | 17 |
| Refusal and agent-reliability scores per model | no | yes, 32 ranked |
| Closed frontier models (GPT, Claude, Gemini) | yes | no |
| Account needed | yes | no, a prepaid key is the account |
| API | OpenAI compatible | OpenAI compatible |
What OpenRouter has
Every uncensored or roleplay model on OpenRouter.
All of them are also on OpenUncensored, under the id in the last column.
| Model | Kind | OpenUncensored id |
|---|---|---|
| Dolphin Mistral 24B Venice Editioncognitivecomputations/dolphin-mistral-24b-venice-edition | Uncensored | venice-uncensored |
| Hermes 4 405Bnousresearch/hermes-4-405b | Roleplay / general | hermes-4-405b |
| Magnum v4 72Banthracite-org/magnum-v4-72b | Roleplay | magnum-v4-72b |
| Euryale 70B v2.2sao10k/l3.1-euryale-70b | Roleplay | l3.1-euryale-70b |
| Cydonia 24B v4.1thedrummer/cydonia-24b-v4.1 | Roleplay | cydonia-24b-v4.1 |
| Skyfall 36B v2thedrummer/skyfall-36b-v2 | Roleplay | skyfall-36b-v2 |
| UnslopNemo 12Bthedrummer/unslopnemo-12b | Roleplay | unslopnemo-12b |
| MythoMax 13Bgryphe/mythomax-l2-13b | Roleplay | mythomax-l2-13b |
| ReMM SLERP 13Bundi95/remm-slerp-l2-13b | Roleplay | remm-slerp-l2-13b |
| Lunaris 8Bsao10k/l3-lunaris-8b | Roleplay | l3-lunaris-8b |
| Weavermancer/weaver | Roleplay | weaver |
What it does not have
Abliterated models OpenRouter does not list.
The uncensored versions of today's strongest open models: DeepSeek V4.1 Flash, GLM 5.3, Qwen 3.8 27B, Qwen3.6 35B A3B, Gemma 4, MiMo V2.6 Flash. They are built by abliteration (removing the direction in the model's weights that drives refusals), heretic tuning, or fine-tuning, and are served by hosts such as Venice, NanoGPT and abliteration.ai, which OpenUncensored routes across with automatic fallback.
See them all on the models page, or ranked by measured refusals, agent reliability and speed on the rankings.
Which to use
When each is the better choice.
- OpenRouter if you need closed frontier models (GPT, Claude, Gemini) and hundreds of mainstream open models in one place, and the occasional roleplay model is enough.
- OpenUncensored if your work needs models that do not refuse: security research, red teaming, fiction and roleplay, medical or legal questions, or agents that should not stop mid task. You also get measured refusal and agent scores to pick a model with.
- Both: many people keep OpenRouter for frontier models and use OpenUncensored for the uncensored ones. Same SDK, different base URL.
Switching
Two lines to change.
Swap the base URL for https://openuncensored.com/api/v1 and the key for an OpenUncensored key. Keys are bought on the credits page with no account.
from openai import OpenAI
# was: OpenAI(base_url="https://openrouter.ai/api/v1", api_key="sk-or-...")
client = OpenAI(base_url="https://openuncensored.com/api/v1", api_key="ou-...")
r = client.chat.completions.create(
model="deepseek-v4.1-flash-1017-uncensored",
messages=[{"role": "user", "content": "Hello"}],
)
print(r.choices[0].message.content)FAQ
Questions people ask.
Does OpenRouter have uncensored models?
A few. Of the 469 models OpenRouter listed on November 10, 2026, one is explicitly uncensored (Dolphin Mistral 24B Venice Edition) and about ten are roleplay fine-tunes such as Hermes 4 405B, Magnum v4 and Euryale. It does not list the abliterated or heretic versions of DeepSeek, GLM, Qwen or Gemma.
Is there an OpenRouter for uncensored models?
OpenUncensored is built as one: one OpenAI compatible API and one key for 44 uncensored and abliterated text models routed across several hosts, plus 17 uncensored image models, with every model scored on refusals and agent reliability.
Can I use OpenRouter's roleplay models through OpenUncensored?
Yes. Every model in the table below is also on OpenUncensored under the id shown, on the same key as the rest of the catalog.
How do I switch from OpenRouter?
Change the base URL to https://openuncensored.com/api/v1, use an OpenUncensored key and pick a model id from the models page. Both are OpenAI compatible, so the rest of your code stays the same.