We went through every hosted service we could find that offers abliterated or uncensored language models. For each one we read the pricing page, model list, privacy policy, and terms, and queried the public model API wherever there was one.
37 providers and more than 500 models made the index. This post maps them, and every model is searchable on the Models page.
Benchmark scores are still to come. Everything below is what providers publish about themselves, checked on October 11, 2026. RedModel was added on October 26, after its developer page stopped describing the API as a demo. Prices are USD per million tokens, input / output.
Five kinds of provider
- Big uncensored catalogs that host hundreds of community checkpoints.
- Branded uncensored services running their own models.
- General inference hosts that don’t market “uncensored” but serve the models anyway.
- Gateways and resellers that route to someone else’s GPUs, often with a markup.
- Writing and roleplay platforms with their own unfiltered fine-tunes and an API on the side.
1. Big catalogs: maximum choice
NanoGPT lists the most uncensored and roleplay models we found: 139 of them. They range from Heretic Gemma 4 and abliterated GLM-5.3 Flash to classic 70B roleplay merges. No account is needed, and you can pay from $0.10 in crypto.
Arli AI has 131 models on flat-rate plans from $10/mo. Most are fine-tunes built on its own Derestricted (abliterated) Qwen3.5 27B, plus Gemma 4 and Qwen 3.8 roleplay and writing tunes.
Featherless.ai fronts the whole Hugging Face long tail. About 1,000 matching checkpoints are available to its API, and we list the 25 most popular. Its cheapest is an abliterated Llama 3.1 8B at $0.05 / $0.08, about $0.09 per 1,000 chats, though per-token prices need its Developer plan (from $50/mo). Only a few 8B roleplay fine-tunes elsewhere cost less; the cheapest is Llama 3 8B Lunaris on OpenRouter at $0.04 / $0.05. Every model is capped at 32K context.
2. Branded uncensored services
Venice.ai is the best-known consumer brand. It runs its own Dolphin-based fine-tune, plus Heretic and uncensored variants of Gemma 4, Qwen, and GLM. Its cheapest is GLM 4.7 Flash Heretic at $0.07 / $0.40. Exactly one of its uncensored models runs end-to-end encrypted.
Abliteration.ai sells its own abliterated GLM-based models from $1 / $3 to $3 / $5, aimed at red-teaming and trust & safety work. Many others resell them.
The rest of this tier is murkier:
- Unjailed AI: GLM-5.3 and DeepSeek variants with no-email signup and Monero; the operator isn’t named.
- NoLimitGPT: in-house model names, with no base models disclosed.
- Uncensored.chat: two undisclosed models, crypto only.
- NoTrack: one undisclosed model; free chat, and an API at $0.15 / $0.50.
- NoFilterGPT: one “custom GPT-architecture” model.
- HackAIGC: mostly image and video, with chat on the side.
- FreedomGPT: its own “Liberty” models plus mainstream ones.
- RedModel: a free research preview of “RedV1”, a red-teaming model for authorized security testing. It doesn’t name the base model.
3. General hosts that quietly serve them
Several mainstream inference clouds host uncensored fine-tunes without marketing them:
- DeepInfra: Hermes 3 70B and 405B, plus Euryale. It’s the upstream behind much of OpenRouter.
- Novita: Stheno 8B from $0.05.
- Nebius: Hermes 4 405B.
- Parasail: TheDrummer’s Cydonia and Skyfall.
- NextBit: Euryale on its own hardware in Spain.
These are often the most transparent options: named companies, clear policies, and published prices. Their selection is smaller and leans toward roleplay fine-tunes rather than abliterated models.
4. Gateways and resellers: check the markup
| Gateway | What you’re actually paying for |
|---|---|
| OpenRouter | Upstream prices with no markup (5.5% card fee on credits); every request names its host |
| Hugging Face | No markup; mostly routes to Featherless, Novita, and DeepInfra |
| Requesty | Upstream + 5% |
| PayPerQ | About 5.5% over upstream; resells Venice models, and takes Lightning and Monero with no account |
| AI/ML API | Prices are almost exactly 1.3754× OpenRouter’s, with no disclosure |
| Electron Hub | Hundreds of Derestricted and Heretic models; some prices are implausibly low |
| DemonRoute | Several times market rate on many models; unnamed upstreams |
If a gateway sells a model you can buy directly, compare prices on the Models page. Search the model’s name and every host shows up side by side.
5. Writing and roleplay platforms
These run their own unfiltered fine-tunes and expose an API, but they’re built for fiction and characters rather than general chat:
- NovelAI: flat rate from $10/mo. Stories are encrypted client-side, and its top model is a GLM-4.6 fine-tune.
- DreamGen: its own Lucid models from $7.83/mo.
- Chub AI: a $20/mo crypto subscription.
- Agnai: Patreon tiers from $5.
- AionLabs: unmoderated roleplay models, also resold by OpenRouter, NanoGPT, and Venice.
- MegaNova: several community models free.
- Infermatic and Mancer: the long-running SillyTavern favorites.
Flat-rate for heavy users
If you chat all day, per-token pricing adds up. These sell unlimited or near-unlimited plans:
| Provider | From | What you get |
|---|---|---|
| Agnai | $5/mo | Roleplay models; bigger ones need higher tiers |
| DreamGen | $7.83/mo | Lucid models, credit-based |
| Infermatic | $9/mo | Curated creative fine-tunes up to 72B |
| Arli AI | $10/mo | All but one of its 131 models (up to 31B, 16K context); the 355B one needs $15+ |
| NovelAI | $10/mo | Unlimited text on its story models |
| Venice.ai | $18/mo | Pro chat across its catalog |
| Chub AI | $20/mo | Unlimited access to its own models |
| Featherless | $25/mo | Unlimited chat across its catalog, 32K context |
Red flags we found
- Unfil.ai: its legal pages still say “localhost:3000”, and its catalog looks stale.
- Unjailed AI and NoLimitGPT: both sell renamed frontier models (Claude, GPT, Grok) far below list price, with no explanation.
- AI/ML API: an apparently undisclosed OpenRouter reseller at a ~37.5% markup.
- Electron Hub: sells Hermes 4 405B at a fraction of every primary host’s price ($0.20 / $0.80 vs $1 / $3), and names no legal entity.
- DemonRoute: prices several times market rate, placeholder-looking uptime stats, and no privacy policy page.
- NoTrack: its privacy policy and API page disagree on whether messages are kept.
- FreedomGPT: its privacy policy says users have “no expectation of privacy”.
- AI Horde: free and open source, but volunteers can see your prompts.
- RedModel: its terms give it a perpetual license to keep and train on everything you send, prompts go to unnamed third-party providers, and no company is named.
Gone since last year
Hyperbolic, Lambda, Kluster, and Targon have all shut down their inference APIs. Together retired its Hermes and MythoMax models. Nous Research’s own portal no longer serves Hermes. The market moves fast, which is why every provider page shows the date its data was checked.
What’s next
Our compliance, capability, and truthfulness runs will turn this map into a ranking. They’ll show which endpoints actually answer, how much intelligence each one keeps, and whether the model you pay for is the model you get. Follow the Signal RSS, or read the methodology.
Spotted an error or a provider we missed? Email [email protected] with a source link.