35 abliterated models, by API.
Every abliterated LLM UncensoredLLM serves as of 2026-11-09: the stock model with its refusal direction removed from the weights, called through the OpenAI SDK on one key and billed per token. The table lists each build’s price and context window from the live catalog; each name opens its own page with the client settings.
Today
- Abliterated builds
- 35
- Lowest output price
- $1.071 / M, Gemma 4 31B Cognitive Unshackled
- Widest context
- 1,048,576 tokens, Abliterated Large Flash
- Passed the daily check
- 34 of 35
- Catalog read
- 2026-11-09
Call one
curl https://api.uncensoredllm.ai/v1/chat/completions \ -H "Authorization: Bearer sk-ullm-live-…" \ -H "Content-Type: application/json" \ -d '{"model": "glm-5.3-abliterated", "messages": [{"role": "user", "content": "Hello"}]}'Every abliterated model
Abliterated models by family
| Family | Abliterated builds | Lowest output $/M | Widest context | Builds |
|---|---|---|---|---|
| DeepSeek abliterated | 9 | $1.12 | 1000K | DeepSeek V4.1 Flash Abliterated, DeepSeek R1 Distill Llama 70B Abliterated, DeepSeek R1 Distill Qwen 32B Abliterated and 6 more |
| Qwen abliterated | 7 | $1.071 | 524K | Qwen 2.5 32B Abliterated, Qwen3.5 27B Blossom V6.4 Derestricted, Qwen3.5 27B BlueStar v3 Derestricted and 4 more |
| Gemma abliterated | 6 | $1.071 | 262K | Gemma 31B Heretic Uncensored, Gemma 4 26B-A4B ChimeraX Heretic, Gemma 4 31B Cognitive Unshackled and 3 more |
| GLM abliterated | 4 | $1.94 | 1049K | GLM-5.3 Abliterated, GLM-5.3 Flash Abliterated, GLM-4.6 Derestricted v5 and 1 more |
| Llama abliterated | 2 | $2.80 | 33K | Llama 3 70B Abliterated v3.5, Llama 3.3 70B Abliterated |
What abliterated means
An abliterated build is the stock model with its refusal direction removed from the weights. Nothing is retrained and no system prompt is added, so it answers prompts the stock model declines and there is no jailbreak prompt to keep alive. Abliterated vs uncensored covers what each edit removes, and jailbroken AI vs abliterated models covers the prompt route against the weights route.
Run an abliterated model without a GPU
Nothing is downloaded and no card is needed: the builds run on the API. Create a key in the console, set the base URL to https://api.uncensoredllm.ai/v1, and send the model id from the table as model. Any client that speaks the OpenAI chat format works; the docs page and the setup guides have the fields for each.
Each build is billed to the token from a prepaid balance. The daily real-model check sends a fixed prompt to every model each day and records whether the answer still matches, so the build behind an id is the one on its label. The leaderboard ranks the abliterated builds by that check and then by output price.
Abliterated models: questions
UncensoredLLM serves 35 abliterated models on one OpenAI-compatible key as of 2026-11-09: GLM-5.3 Abliterated, GLM-5.3 Flash Abliterated, DeepSeek V4.1 Flash Abliterated, Abliterated Large Flash, Abliterated Model, Abliterated Model Large and 29 more. The base URL is https://api.uncensoredllm.ai/v1, the model id goes in the model field, and each request is billed per token from a prepaid balance that starts at $10.
The stock model with its refusal direction removed from the weights. Nothing is retrained and no prompt is added: the direction the model uses to decline a request is projected out, so it answers prompts the stock model declines while the rest of its behaviour stays close to the original. See abliterated vs uncensored.
From $1.071 input and $1.071 output per million tokens on Gemma 4 31B Cognitive Unshackled, the lowest-priced abliterated build on 2026-11-09. Every build has its own rate in the table, there is no subscription, and a prepaid balance from $10 pays per token. See the price list.
Abliterated Large Flash, at 1,048,576 tokens, is the widest of the 35 abliterated builds here.
No. An abliterated build is an edit of the stock weights with nothing retrained; an uncensored build is a fine-tune trained to answer, or an official model served without a filter. UncensoredLLM lists 35 abliterated builds among 172 models, and labels each one in the Build column. See every build by family.
No. The change is in the weights, so there is no jailbreak preamble to write, keep alive or pay for on every message. Send the request as you would to any OpenAI-format model. See jailbroken AI vs abliterated models.
No. An Ollama or GGUF copy runs on your own machine and needs the memory for it. The builds here run behind the API and bill per token: set the base URL, paste a key and send the model id. See the Ollama tags next to the API builds.
Any client that speaks the OpenAI chat format: the OpenAI SDK, SillyTavern, Cursor, Cline, Open WebUI and LangChain among them. Point the client at https://api.uncensoredllm.ai/v1, paste the key and type the model id from the table. See the setup guides.