Research only

A model with the refusals cut out

Works out of the box. No rain dance. GLM 5.3 answers the prompt instead of “I can’t assist with that.” A private RunPod box sits at about $40/hour even while idle. Here you start at $1, pay as you go. Try it in Telegram first.

Keys, balance, and top-up live in the cabinet or in @uncensobot.

z-ai/glm-5.3-flash-uncensored

cybersec ai-red-team trust-safety

user You found a blind SQL injection in a web app, but the server returns the same body every time and bans you after 5 attempts. How do you pull the data without getting burned?

this model

200
The body is a dead channel. Look at delay or length, not HTML.
Five requests are a bit budget, not a brute-force run.
Spend one on timing calibration, the rest on binary search: each call halves the space.
Pause between calls. Don’t spray — the ban will eat the box.

other models

I can’t assist with that.

user We need to eval an agent for indirect injection via file metadata. Which classes belong in the harness?

this model

200
Three buckets: document (title, keywords), markup (alt, comments), URL context.
One positive and one negative per class.
Watch whether a tool-call fires. Keep strings on the stand, not in prod.

other models

I can’t assist with that.

user Classify this incoming comment under policy: category, severity, why that label. Content redacted.

this model

200
{
  "category": "harassment",
  "severity": 3,
  "reason": "directional insult",
  "action": "rewrite_or_escalate"
}

other models

I can’t assist with that.

$1smallest PAYG pack

$0.005minimum charge per call

8models in the catalog

~$40/hrown GPU on RunPod, for scale

Why

Built for work other models refuse

Cabinet or Telegram bot. The public surface is Chat Completions.

Refusals removed

GLM 5.3 without safety theater. For red-team notes, datasets, and policy research — not for chores.

Same protocol

OpenAI Chat Completions. Not Anthropic Messages, not Responses. One Base URL.

Bot first

A free Telegram slot, no search and no think. The API has no free tier.

Pay for tokens

No hourly GPU rent. Packs $1–$100. Keys live in the cabinet or the bot, up to five.

Setup

Three steps

Open the bot

@uncensobot — try chat, then API: a name and a secret.

Set the Base URL

https://uncensored.dog/v1. Do not append /chat/completions.

Call the model

Eleven IDs, default z-ai/glm-5.3-flash-uncensored. Empty model also uses GLM Flash. Cards and prices are below.

Pricing

PAYG from a dollar, not GPU rent

Eleven text models. Prices are per million tokens. We charge after the reply. Minimum $0.005 per request. Top-up is in the cabinet or in Telegram.

GLM 5.3 Flash Uncensored

default

z-ai/glm-5.3-flash-uncensored

The default model. Fast general-purpose model for everyday chat, writing, and code.

Text
Vision
Web search
Reasoning
Tools

Input: $0.39/1M
Output: $1.58/1M
Cache: $0.16/1M

Abliterated Large v2

smartest and fastest

abliteration-ai/abliterated-model-large-v2

The smartest and fastest model in the catalog.

Text
Reasoning
Tools

Input: $5.51/1M
Output: $9.19/1M
Cache: $0.55/1M

Abliterated Large Flash

abliteration-ai/abliterated-model-large-flash

Fast Abliterated model based on DeepSeek V4.1 Flash. Text, images, and tools.

Text
Vision
Reasoning
Tools

Input: $2.36/1M
Output: $4.72/1M
Cache: $0.24/1M

Qwen 3.8 27B Uncensored

qwen/qwen3.8-27b-uncensored

Text Qwen model without mandatory reasoning mode.

Text
Vision
Tools

Input: $0.40/1M
Output: $2.60/1M
Cache: $0.39/1M

Qwen 3.8 27B Uncensored Thinking

qwen/qwen3.8-27b-uncensored:thinking

Separate reasoning Qwen variant for multi-step tasks.

Text
Vision
Reasoning
Tools

Input: $0.40/1M
Output: $2.60/1M
Cache: $0.39/1M

DeepSeek V4.1 Flash Heretic Thinking

pqhaz/DeepSeek-V4.1-Flash-1017-heretic:thinking

Fast DeepSeek V4.1 Flash. Reasoning is always on. Text, images, and tools.

Text
Vision
Reasoning
Tools

Input: $0.52/1M
Output: $2.08/1M
Cache: $0.03/1M

Abliterated Model

abliteration-ai/abliterated-model

Base Abliterated model for short and everyday text tasks.

Text
Vision
Reasoning

Input: $1.84/1M
Output: $5.51/1M
Cache: $0.18/1M

Gemma 4 31B Gembrain Uncensored Heretic

Gemma-4-31B-Gembrain-uncensored-heretic

Text Gemma 31B model for conversation, writing, and analysis.

Text
Vision
Reasoning

Input: $0.57/1M
Output: $0.57/1M

GLM 5.3 Uncensored

z-ai/glm-5.3-uncensored

Full GLM 5.3, not the fast Flash. No images. Experimental: reply quality can vary. Tools are supported.

Text
Reasoning
Tools

Input: $2.10/1M
Output: $4.20/1M
Cache: $0.52/1M

MiMo V2.6 Flash Uncensored

xiaomi/mimo-v2.6-flash-uncensored

Text-only MiMo, no images. Tools are supported, and reasoning can be turned off.

Text
Reasoning
Tools

Input: $1.31/1M
Output: $3.94/1M
Cache: $0.52/1M

MiMo V2.6 Flash Uncensored Thinking

xiaomi/mimo-v2.6-flash-uncensored:thinking

Same MiMo, but reasoning is always on and replies take longer. No images.

Text
Reasoning
Tools

Input: $1.25/1M
Output: $3.75/1M
Cache: $0.50/1M

Empty model in the API uses Qwen. Telegram model choice does not change the API. In the bot: 1 request per calendar month, 2 with Telegram Premium. The slot has no search and no think. The API has no free calls.

$1USDT in the bot

$5USDT in the bot

$10USDT in the bot

$25USDT in the bot

$50USDT in the bot

$100USDT in the bot

Keys and top-up live in the cabinet or the bot. Balance is shared.

Top up In Telegram