This list of uncensored AI tools is ordered by the task on a security engagement rather than by vendor: a browser chat, the API, coding agents, scoring, the paperwork, the budget, media, local weights, and two hosted apps described from their own pages. Eight of the ten run on a single prepaid key, and the prices below come from the live catalog on every build of this page. Catalog of 2026-11-17.
Ten tools, ordered by what a tester is doing
Most pages that rank for this phrase line tools up by vendor. This one lines them up by the task on the engagement, since one model serves a different purpose in each. Eight of the ten run on a single prepaid key here. The other two are hosted apps that also rank for the phrase; their rows come from their own pricing pages as read on October 27, 2026.
| # | Task | Tool | What it gives you |
|---|---|---|---|
| 1 | Ask a question in a browser tab | Playground | Free turns without an account on the lower-priced builds; a key for the rest |
| 2 | Your own scripts and agents | The API | One base URL, one prepaid key, OpenAI and Anthropic wire formats, billed per token |
| 3 | Code review and exploit review in a coding agent | Code builds in OpenCode, Claude Code or Cursor | The code builds configured as the agent model; the security builds on the red teaming page |
| 4 | Scoring a finding | CVSS 3.1 calculator | Eight base metrics in, base score, severity and vector string out; no account |
| 5 | Paperwork before the test starts | Rules of engagement builder | Scope, timing, methods and an authorization letter assembled in the browser, then copied or printed |
| 6 | Budgeting the model spend | Engagement cost estimator | What an engagement costs per request, per day and in total at the catalog rates |
| 7 | Image and video generation | Media models | 10 image models priced per image and 3 video models priced per second |
| 8 | Weights on your own hardware | Ollama, LM Studio | GGUF files of the same builds; your GPU, nothing billed per token |
| 9 | A hosted private chat app | Venice | Free tier with daily prompt limits, paid plans by the month (read October 27, 2026) |
| 10 | Any open model, hosted | Featherless | Monthly flat plans across a very large open-weight catalog (read October 27, 2026) |
Where a chat build is called uncensored on this site, the word has one meaning: the refusal has been taken out of the weights, either by abliteration or by a fine-tune on refusal-free data. Abliterated vs uncensored compares the two methods, and what is abliteration explains the first.
The per-token rates under the tools
Three builds stand in for the catalog, one of each kind, read live when this page was built: the assistant that heads the models page, a community fine-tune from the far end of the list, and one of the security builds. Each tool that runs on the key is billed at rates like these.
| GLM-5.3 Uncensored | Uncensored | $0.282 | $0.846 |
| Cydonia 24B v4.3 | Roleplay | $0.484 | $0.60 |
| GLM-5.3 Flash Cybersecurity | Security | $0.5093 | $1.6975 |
A daily probe at /proof confirms that the id you call still answers as the build on its label. On November 17, 2026, 143 of the 170 listed models passed. Credit starts at $10, is drawn down per token and never expires.
1. A browser tab: the playground
The playground puts every per-token build behind one prompt box. There is nothing to install and no hidden system prompt, and when one build hedges on a question about a CVE you are triaging, the picker moves the same question to another build. The lower-priced builds answer the first turns of the day without a key; the home page has the same demo turns with no account at all. For the "AI chatbot with no restrictions" search this is the answer, as long as the model matters more than the chat window.
2. The API: one key, every model
For a recon script, a triage pipeline or a custom agent. Point the OpenAI SDK, in any language, at two changed values:
- Base URL
- https://api.pentestllm.ai/v1
- Key
- sk-ptl-live-…Issued once in the console; copy it then
- Model
- glm-5.3-uncensoredOr any id on the models page
Each response reports the settled cost of that call in its headers, and the Anthropic message format is accepted on the same key. Open WebUI, LiteLLM, n8n and a Discord bot each have a guide; Python and Node samples are on the docs page.
3. Coding agents: the code builds on an engagement
The code builds (uncensored-coding-agent, glm-5.3-flash-code-uncensored) return code when asked to review a proof of concept against a host in scope, unpack a sample or explain a payload, instead of declining. They run in OpenCode, Claude Code, Cursor and Open WebUI; the uncensored coding AI guide configures each and prices a working day from the catalog. The 9 security builds, and the agent settings for them, are on red teaming. Authorization in writing for every target is still the rule of the acceptable-use page.
4. Scoring: the CVSS 3.1 calculator
The CVSS 3.1 calculator takes the eight base metrics and returns the base score, the severity band and the vector string. Paste a vector from a vendor advisory to check its score, or send a scored finding to a colleague by link. It needs no key and no account, and it is the quickest way to put a defensible number next to a finding before the report is drafted.
5. Paperwork: the rules of engagement builder
The rules of engagement builder assembles the document a client and a tester sign before anything is touched: what is in scope, when testing may run, which methods are allowed and who gets the call when something breaks. It also produces the authorization letter. Fill it in the browser, then copy, download or print. Every model on this site is for systems covered by a document like that one.
6. Budget: the engagement cost estimator
The engagement cost estimator turns requests per day, tokens per request, a day count and a build into a figure per request, per day and per engagement, at the live per-token prices. Agent loops that resend long tool output are where the money goes, so the input rate matters most; the estimator makes that visible before the first top-up. The pricing page holds the same rates with the tool fees.
7. Media: image and video models on the same key
Open diffusion and video models, billed per image or per second from the same balance. Two rules apply to every media prompt: nothing that shows a minor, and no edit of a real person’s photo without their consent. The acceptable use policy has the wording. The media catalog lists each tier and the request shape.
| Per | Endpoints | |||
|---|---|---|---|---|
| LTX-2.3 Image-to-Video Uncensored | Video | $0.08 | second | |
| Wan 3.0 Image-to-Video Uncensored | Video | $0.20 | second | |
| Wan 3.0 Video Uncensored | Video | $0.20 | second | |
| Anime Image Uncensored | Images | $0.03 | image | |
| Chroma Uncensored | Images | $0.08 | image | |
| CyberRealistic XL Uncensored | Images | $0.0136 | image | |
| Qwen Edit Uncensored | Images | $0.08 | image | |
| Qwen Image 2.1 Uncensored | Images | $0.0375 | image | |
| Qwen Image 3 Uncensored | Images | $0.08 | image | |
| Seedream 5.0 Lite Edit Uncensored | Images | $0.0313 | image | |
| Vision Image Uncensored | Images | $0.04 | image | |
| Vision Pro Image Uncensored | Images | $0.09 | image | |
| WAI Illustrious SDXL Uncensored | Images | $0.0136 | image |
8. Local: the same GGUF files in Ollama or LM Studio
Most of the builds on the key are public weights with GGUF files on Hugging Face, which is why the "uncensored AI github" search and the "run it offline" search land on the same models. On your own GPU there is no per-token bill and the prompts stay on the box, which matters when a client forbids data leaving their network. The Ollama guide walks through the download; LM Studio opens the same files.
9 and 10. Two hosted apps, in their own figures
Venice is a private chat app with an API attached. On October 27, 2026 its pricing page showed a $0 Free plan on base models with 10 text prompts and 15 image prompts a day, Pro at $18 a month with unlimited text prompts, 1,000 images a day and 100 credits a month, then Pro Plus at $68 and Max at $200 with API spend included. It wins on setup time for private chat. The vendor chooses the model, and the bill is monthly. The Venice comparison prices one month on each side.
Featherless serves open models from Hugging Face on flat monthly plans. Its pricing page on October 27, 2026 showed Chat at $25 a month with 4 concurrent units, Developer at $50 a month in credits with 100 concurrent units, and a Business tier, over what it counted as 40,000+ open models. It wins on breadth. You pay for a plan and for concurrency, not per token.
Safe for the tester, safe for everyone else
"Is uncensored AI safe" is two questions. For you: the API does not store the request body unless the key has opted into the 7-day replay, there is no browsable account history, and every key carries its own spend cap. For everyone else: the model writes what it is asked to, so the limits live in the acceptable use policy: nothing involving a minor, no real person without consent, no action against a system you do not own or hold written permission to test. On the image and video models one check in the API rejects any prompt describing a minor before it is priced.
A build with its refusal direction removed still says no once in a while. Each such turn is refunded, up to $5 a day per account. Keys come from the console, prepaid from $10, with no subscription required.
Questions
Match the tool to the task: the playground for a quick question, the API for scripts, the code builds in OpenCode, Claude Code or Cursor for review work, the CVSS calculator for scoring, the rules of engagement builder for the paperwork, the cost estimator for the budget, the media models for images and video, and Ollama or LM Studio to run the weights offline. Every model behind them has the refusal removed from the weights; 170 are in the catalog today. See the models page with prices.
GitHub carries the runners and the abliteration code (llama.cpp, Ollama, Open WebUI and the like); the weights themselves sit on Hugging Face. The same ids answer over the API with no download. See builds by family.
The model follows the prompt, so the boundaries are the acceptable use policy: nothing involving a minor, no real person without consent, no system you lack written authorization to test. Image and video prompts that describe a minor are rejected by one check in the API before pricing. Request bodies are not retained unless a key opts in. See the acceptable use policy.
Yes. The code builds answer exploit-review and malware-analysis questions with code, and run inside OpenCode, Claude Code, Cursor and Open WebUI. The coding guide covers the setup; the red teaming page covers the security builds. See the uncensored coding AI guide.
glm-5.3-uncensored heads the models page and is the playground default; cydonia-24b-v4.3 is a community fiction tune at the other end of the catalog. The models page lists each build with its price, context window and the verdict of the daily check. See what "no restrictions" means in practice.
Venice describes itself as private, uncensored chat with no card needed on its free plan (its pages, read October 27, 2026). As with any hosted app, the vendor picks the model and the filter; over an API those decisions are yours. See Venice compared.
Yes. The home page prompt box and the playground give a few free turns with no account. The API needs one key, created once; nothing asks you to log in again after that. See the playground.
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.