Does OpenRouter have uncensored models? What it lists, and where else to get them
The Morse teamPublished Facts checked 7 min read
Yes, but only a few. Of the 476 text models in OpenRouter's model list on November 20, 2026, two describe themselves as uncensored, Venice: Uncensored and TheDrummer's Cydonia 24B V4.1, and a third, Nous Research's Hermes 4 405B, says it's trained for "lower refusal rates." None of the three is free, and each is served by a single provider.
If you want an uncensored model by API with privacy you can check, Morse Private Inference serves DeepSeek V4.1 Flash 1017 Uncensored at $0.38 per million input tokens and $1.50 per million output tokens. It runs only at E2EE, the most private of Morse's four tiers, so every message is encrypted to a key held only by the model's attested enclave and Morse relays ciphertext it can't read.
To build this list, we downloaded OpenRouter's full model list on November 20, 2026, searched every model's id, name, and full description for "uncensored," "abliterated," "unfiltered," "dolphin," "venice," and "hermes," and read the matching model pages, OpenRouter's FAQ, terms, data collection and zero data retention docs, and provider logging table. We read Venice's, Featherless.ai's, and NanoGPT's own model lists and docs the same day. We sent no requests through any of them, so every figure below is the company's own, listed under Sources. OpenRouter deserves credit for how much it publishes: each model page names its provider and that provider's data policy, and the model list says which models carry moderation.
Morse, which publishes this page, makes a private AI chat app with GPT, Claude, Gemini, DeepSeek, and uncensored models, encrypted so even Morse can't read your chats. For developers it runs Morse Private Inference, a pay-as-you-go API for 80+ open-weight models with no subscription, and Morse keeps no prompt and no reply on any tier. Anyone can run npx @askmorse/verify inference.trymorse.ai to check the open-source Morse Enclave that every request passes through.
Uncensored models by API, side by side
As of November 20, 2026. Morse first, then OpenRouter's three in its own order of price, then the other APIs. Prices are US dollars per million input and output tokens, from each company's own list.
| Model | Where | Input / output | Context | Who can read the prompt |
|---|---|---|---|---|
| DeepSeek V4.1 Flash 1017 Uncensored | Morse Private Inference | $0.38 / $1.50 | 1M | Only the model's attested enclave (E2EE) |
| Venice: Uncensored | OpenRouter, served by Venice | $0.20 / $0.90 | 128K | Venice, listed as zero retention |
| Cydonia 24B V4.1 | OpenRouter, served by Parasail | $0.30 / $0.50 | 131K | Parasail, listed as zero retention |
| Hermes 4 405B | OpenRouter, served by Nebius Token Factory | $1.00 / $3.00 | 131K | Nebius Token Factory, listed as zero retention |
| Venice Uncensored 1.2 | Venice API | $0.20 / $0.90 | 128K | Venice, with zero data retention |
| 1,588 models tagged uncensored | Featherless.ai | From the Developer plan, per token | Up to 256K | Featherless.ai, which says it doesn't log prompts |
Which uncensored models does OpenRouter list?
OpenRouter's text list held 476 models on November 20, 2026, and 739 across every output type. Three match the search, and nothing in the list uses the words "abliterated" or "unfiltered."
- Venice: Uncensored (
cognitivecomputations/dolphin-mistral-24b-venice-edition). OpenRouter's page calls it "an 'uncensored' instruct-tuned LLM" that removes "default safety and alignment layers typically found in mainstream assistant models." Venice is its only provider, and the page says "OpenRouter forwards every request without routing." - Cydonia 24B V4.1 (
thedrummer/cydonia-24b-v4.1). Its description reads "Uncensored and creative writing model based on Mistral Small 3.2 24B," served by Parasail alone. - Hermes 4 405B (
nousresearch/hermes-4-405b). It doesn't use the word uncensored. Its page says it is "trained for steerability, lower refusal rates, and alignment toward neutral, user-directed behavior," served by Nebius Token Factory.
All 12 free variants in the text list are mainstream models, so there's no free uncensored model on OpenRouter today. OpenRouter's pricing page adds a 5.5% fee when you buy credits on its Standard plan.
What does OpenRouter moderate?
OpenRouter applies moderation model by model. Its model list marks 144 of its 476 text models with is_moderated: true, which its models guide defines as "Whether content moderation is applied," and most of them come from OpenAI (88) and Anthropic (31). None of the three models above is marked.
A request that trips moderation gets a 403, which OpenRouter's errors page lists as a "guardrail block, or moderation flag," with provider_name naming the provider that asked for it. Its terms, last updated August 31, 2026, also say OpenRouter may "screen, remove, edit, or block any Inputs that in our sole judgment violates these Terms," and section 7.8 bars content "not in compliance with the Terms of Service for the Model or Provider you are using." Its data collection page says OpenRouter screens inference requests, zero data retention traffic included, for child sexual abuse material.
How private are OpenRouter's uncensored models?
OpenRouter's provider logging table lists Venice, Parasail, and Nebius Token Factory as "Zero retention" and "Does not train," and each model page shows the same policy. OpenRouter's FAQ says "Prompt and completion are not logged by default," and its data collection page adds that it "samples a small number of prompts for categorization," stored anonymously unless you opt in to logging, plus metadata such as token counts and latency for each request. Our guide to Morse vs OpenRouter goes through those settings one by one.
Each of the three runs on one provider, so the provider reads your prompt in the clear and its zero retention is a policy you take on trust. For a model where no provider can read the prompt at all, the request has to be encrypted before it leaves your code.
Where else can you get uncensored models by API?
Morse Private Inference
Morse Private Inference serves DeepSeek V4.1 Flash 1017 Uncensored, an abliterated version of DeepSeek V4.1 Flash 1017, under the id deepseek-ai/deepseek-v4.1-flash-1017-abliterated. It reasons, takes a context of up to 1M tokens, and costs $0.38 per million input tokens and $1.50 per million output tokens, paid from credits with no subscription. Your first top-up gets 50% extra.
It runs only at E2EE, which works as a privacy floor nothing can lower. Your client fetches the model enclave's attestation from GET /v1/attestation, checks it, and encrypts each message to the enclave's key, and Morse relays the ciphertext without being able to read it. The reply comes back encrypted to your key, with a signed receipt over the request and the reply. The end-to-end encryption guide walks through each step, and the models page lists the price.
Venice
Venice's public model list flags 17 of its 148 text models as uncensored on November 20, 2026, through an OpenAI-compatible API at https://api.venice.ai/api/v1. Venice Uncensored 1.2 costs $0.20 and $0.90 per million tokens, and its docs page says "no prompt data is stored in any capacity." Its default text model, DeepSeek V4.1 Flash 1017 Uncensored, costs $0.33 and $1.32. Several flagged models carry Venice's "anonymized" label, which its methodology page explains: "Proxied to the upstream lab without your identity. The lab may retain prompts."
Featherless.ai
Featherless.ai serves open fine-tunes from Hugging Face through an API its quickstart calls "OpenAI compatible," at https://api.featherless.ai/v1. Its model filters showed 1,588 models tagged uncensored and 1,385 tagged abliterated on November 20, 2026. Its docs say "Featherless does not log chats, prompts, or completions sent through our API." Its $25-a-month Chat plan is "Not for reselling, app/API traffic, background automation," so API use starts on the Developer plan, from $50 a month and billed per token.
NanoGPT
NanoGPT's model list put 51 of its 615 models in an "Uncensored" category on November 20, 2026, behind an API at https://api.nano-gpt.com/api/v1. They include GLM 5.3 Flash Uncensored at $0.20 and $0.80 per million tokens, and an abliterated DeepSeek V4.1 Flash at $0.297 and $1.045.
What do "uncensored" and "abliterated" mean?
An uncensored model answers questions that a mainstream assistant declines, such as frank questions about health, law, money, or testing your own security. "Abliterated" names one way to get there: researchers find the direction in a model's weights that drives refusals and remove it, leaving the rest of the model as it was. Our guides to what an uncensored AI model is and what an abliterated model is explain both in more depth.
Which uncensored API fits your project?
- Choose Morse Private Inference if you want an uncensored model where no one else can read the prompt, Morse included, with an enclave you can check and no prompt or reply kept.
- Choose OpenRouter if you already use it and one of its three models fits, and a provider under zero retention is enough.
- Choose Venice if you want several uncensored models behind one OpenAI-compatible API.
- Choose Featherless.ai if you want to try many open fine-tunes, and a monthly plan suits you.
- Choose NanoGPT if you want a large catalog with an Uncensored category to browse.
How Morse keeps uncensored requests private
- Encrypted end to end. Morse's uncensored model runs only at E2EE, so the text is readable only inside the model's attested enclave.
- Nothing kept. Morse keeps no prompt and no reply, on any tier. For billing it records the model, the token counts, the cost, the time, and the key.
- Not linked to you. A provider sees Morse, never your account, your card, or your address, and Morse drops
user,safety_identifier, andprompt_cache_keybefore a request leaves. - A check you can run.
npx @askmorse/verify inference.trymorse.aichecks the Morse Enclave's AWS Nitro attestation and its open-source code at github.com/trymorse/morse-enclave. Verify it yourself explains each step.
In the Morse chat app, uncensored models come with Pro and Max, with your history stored on your device, encrypted. Read the Morse docs, or make a free account and create your first key.
Morse publishes this comparison. We've tried to keep it fair; if something's wrong or out of date, email hello@trymorse.ai.
OpenRouter is a trademark of OpenRouter, Inc. Morse is not affiliated with or endorsed by OpenRouter, Inc.
Sources
All read November 20, 2026.
- OpenRouter: model list, Venice: Uncensored, Cydonia 24B V4.1, Hermes 4 405B, models guide, errors, FAQ, terms, data collection, provider logging, pricing
- Venice: model list, Venice Uncensored 1.2, model methodology
- Featherless.ai: quickstart, pricing, privacy and logging, uncensored models, abliterated models
- NanoGPT: docs, model list
Common questions
Does OpenRouter have uncensored models?
A few. On November 20, 2026, two of the 476 text models in OpenRouter's list describe themselves as uncensored, Venice: Uncensored and TheDrummer's Cydonia 24B V4.1, and Nous Research's Hermes 4 405B says it's trained for lower refusal rates.
Are any of OpenRouter's uncensored models free?
No. None of the three has a free variant in OpenRouter's list. Venice: Uncensored costs $0.20 per million input tokens and $0.90 per million output tokens, Cydonia 24B V4.1 $0.30 and $0.50, and Hermes 4 405B $1.00 and $3.00.
Does OpenRouter moderate content?
On some models. OpenRouter's model list marks 144 of 476 text models with is_moderated, the field its docs describe as whether content moderation is applied, most of them OpenAI's and Anthropic's. None of the three models above is marked.
Who serves OpenRouter's uncensored models?
One provider each: Venice serves Venice: Uncensored, Parasail serves Cydonia 24B V4.1, and Nebius Token Factory serves Hermes 4 405B. OpenRouter's provider logging table lists all three as zero retention and not training.
Does Morse have an uncensored model in its API?
Yes. Morse Private Inference serves DeepSeek V4.1 Flash 1017 Uncensored, a reasoning model with a 1M-token context, at $0.38 per million input tokens and $1.50 per million output tokens, end-to-end encrypted.
How private is Morse's uncensored model?
It runs only at E2EE, Morse's most private tier. Each message is encrypted to a key only the model's attested enclave holds, so Morse relays ciphertext it can't read, and Morse keeps no prompt and no reply.
What does abliterated mean?
An abliterated model has had the direction in its weights that drives refusals found and removed, so it answers instead of declining. Morse's uncensored DeepSeek model is an abliterated version of DeepSeek V4.1 Flash 1017.
Where else can I get uncensored models by API?
Morse serves one, end-to-end encrypted. Venice's public list flags 17 of its 148 text models as uncensored, Featherless.ai tags 1,588 models uncensored and 1,385 abliterated, and NanoGPT puts 51 of its 615 models in an Uncensored category.
Does OpenRouter keep prompts sent to its uncensored models?
Not by default, its FAQ says, and its provider logging table lists Venice, Parasail, and Nebius Token Factory as zero retention. OpenRouter does keep metadata such as token counts and latency for each request.
Do uncensored models on Morse come with the chat app too?
Yes. In the Morse chat app, uncensored models come with Pro and Max, for frank answers on health, law, money, and security that other apps refuse.