There is no unfiltered ChatGPT, because the filter is the product. There are open-weight models with the refusal cut out of their weights, and you can talk to one in the browser right now or call it through the SDK ChatGPT’s API uses. Think of this post as the ChatGPT chapter of AI with no restrictions: what to open today, why a prompt will not get you there, and the builds at the top of the models page with live prices.
Use this instead of an unfiltered ChatGPT
Two routes, neither of them a prompt trick. The fast one is /chat: choose a build, type, read the answer. Three turns a day are free with no account, no email and no phone number, and a system prompt and temperature live in Settings. The playground runs the same builds on a key. The second route is the API, which takes the three settings ChatGPT’s own API takes, with a different base URL and key:
- Base URL
- https://api.unrestrictable.ai/v1
- Key
- sk-unr-live-…Made in the console; displayed one time
- Model
- glm-5.3-uncensoredGLM-5.3 Uncensored
from openai import OpenAI # Same SDK as ChatGPT's API. Only base_url and api_key differ.client = OpenAI(base_url="https://api.unrestrictable.ai/v1", api_key="sk-unr-live-…")reply = client.chat.completions.create( model="glm-5.3-uncensored", messages=[{"role": "user", "content": "Write the villain's monologue. He wins, and he explains how."}], stream=True,)for chunk in reply: print(chunk.choices[0].delta.content or "", end="")The prices below are per million tokens and were read from the live catalog as this page rendered. They are the 3 per-token builds that the models page lists first, in its order.
| GLM-5.3 Uncensored | Uncensored | $0.282 | $0.846 |
| GLM-5.3 Abliterated | Abliterated | $2.1825 | $4.365 |
| DeepSeek V4 Pro Uncensored | Uncensored | $0.97 | $1.94 |
A single turn of 1,500 tokens in and 500 out costs $0.0008 on glm-5.3-uncensored at today’s rate. A month of 1,000 turns like it comes to $0.85. You top up from $10 by card, WeChat Pay or crypto; a monthly plan is optional. Today's verdict on this build from the daily check: steady since oct 12.
Why an unfiltered ChatGPT cannot exist
ChatGPT is a product. The model under it is served by OpenAI alone, with a system prompt you never see, a moderation layer you cannot turn off, and weights that are not published. The filter is not an option bolted onto ChatGPT; take it away and what is left is not ChatGPT. Every result for this search that offers an unfiltered ChatGPT turns out to be one of three things: another model wearing a ChatGPT-style interface, a chat-platform character called “ChatGPT unfiltered”, or a prompt.
Restrictions sit in up to three places: in the weights, in the system prompt ahead of the model, and in a moderation layer around the whole thing. ChatGPT has all three. Our AI with no restrictions pillar walks through each layer and what removing it means. This post is the ChatGPT branch of that pillar.
Prompts break; edited weights do not
A prompt that talks ChatGPT out of its filter is a jailbreak. It may work the day it is posted and fail the moment the hosted model, or anything wrapped around it, is updated. It also eats tokens on every turn, since the model keeps nothing between requests. A model whose refusal has been removed from the weights needs no prompt. The publisher of the build made the edit once, and nobody can undo it from the outside. In 2024 Arditi and colleagues showed that refusal in open-weight chat models runs along one direction in activation space (their paper). Deleting that direction is abliteration.
What exists: open weights with the refusal edited out
You can edit open weights. You cannot edit a hosted product. Every build on the models page is an open-weight model with the trained refusal taken out, and moving from one to another means changing one string. OpenAI has released open-weight models of its own under the GPT-OSS name; no unfiltered build of them is on this key today. The first builds on the list, as read on 2026-11-16:
glm-5.3-uncensored: GLM-5.3 Uncensored, tagged uncensored; $0.282 in and $0.846 out per million tokens; $0.85 for 1,000 turns a month; daily check verdict steady since oct 12.glm-5.3-abliterated: GLM-5.3 Abliterated, tagged abliterated; $2.1825 in and $4.365 out per million tokens; $5.46 for 1,000 turns a month; daily check verdict steady since oct 30.deepseek-v4-pro-uncensored: DeepSeek V4 Pro Uncensored, tagged uncensored; $0.97 in and $1.94 out per million tokens; $2.43 for 1,000 turns a month; daily check verdict steady since oct 12.
None of these is ChatGPT, and none has ChatGPT’s tools, memory or browsing. What they do have: the chat-completions format that ChatGPT’s API speaks, and weights with the refusal removed. Each model page shows the price, the day’s verdict and the request to send, and the docs page has the OpenAI SDK setup for Python and Node. Abliterated vs uncensored tells the two edits apart.
Free, no login: what the "unfiltered ChatGPT" apps give you, and what differs here
Search for “unfiltered ChatGPT free” or “no login” and what comes back is chat apps, NoFilterGPT, FreedomGPT and uncensored.chat among them, plus a Venice article (Google, read October 27, 2026). A hosted chat app chooses the model for you, along with whatever is placed in front of it, and its owner can tighten the filter whenever they like.
Here the model is your choice, and no layer in between can switch a filter on. /chat is free for three turns a day with no account; a free account raises that to five messages a day, and after that credit is prepaid. The API is paid from the first token, and the first section prices a turn and a month of them. Chats stay in your browser: request bodies are not stored, though usage records are. More on that at private AI chat.
The gateway does not filter chat requests by topic. What is never allowed is on the acceptable-use page. Any image or video prompt that involves a minor is turned down before any price is worked out.
Questions
No. The filter is part of the ChatGPT service, and the model's weights are not public. What you can use is an open-weight model whose refusal was edited out, such as GLM-5.3 Uncensored, in /chat or through the API. See the GLM-5.3 Uncensored page.
You use a different model. In the browser, open /chat: three turns a day with no account. By API, set the base URL https://api.unrestrictable.ai/v1, a key from the console and a model id from the models page (glm-5.3-uncensored, for instance) in the same OpenAI SDK call you would send to ChatGPT. See the browser chat.
Not the product itself. A jailbreak prompt dies whenever the hosted model is updated, and it burns tokens every turn. Real uncensoring is an edit to the weights, which only open-weight models allow, and that edit is what every build here has. See how the edit works.
There is no official one. Anything sold as a "dark ChatGPT" is a different model behind a ChatGPT-style screen, or a prompt. The real thing is an open-weight model whose refusal was removed, and the models page names each one for what it is. See the uncensored model list.
They circulate, and they break when the model is updated. This page does not carry one. A model with the refusal removed from the weights does not need a prompt, and the vendor cannot switch the refusal back on.
Not ChatGPT. Here, /chat gives you three turns a day with no account, email or phone number. After that a free account allows five messages a day, or you top up from $10 and pay per token at the rates in the table. See the pricing page.
OpenAI does not offer one. Our pillar post, AI with no restrictions, explains the three layers a restriction can sit in. This post is its ChatGPT branch, and the answer is the same: an open-weight model whose refusal was removed. See AI with no restrictions.
No. ChatGPT filters its image model as it filters everything else. The media page here lists uncensored image and video models on the same key, priced per image or per second of video. See the uncensored media models.
Every model above is on the uncensored model list with its live price per million tokens. Per-token pricing sets the whole catalog side by side, and the family directory groups the builds by base model.