Red teaming

Uncensored AI for cybersecurity.

7 security builds that answer as an analyst, 5 general uncensored builds you steer yourself. One OpenAI-compatible key, priced per token, in Strix, HackingBuddyGPT and CAI.

An uncensored LLM API for security research: ask for a CVE write-up, exploit triage, a Sigma or YARA detection rule, a malware sample’s behaviour explained, a decompiled function read back, or a finding with its proof of concept for a system you are authorized to test.

Get a key

Picking a model

Cyber Frontier
$44.10 per 1,000 triage turns
triage, CVE write-ups and reports with no system prompt of your own
GLM-5.3 Uncensored
$2.37 per 1,000 triage turns
the budget pick for agent loops that send thousands of turns
Cyber Frontier
1024K context
for a tool that reads long logs or a whole repository

A triage turn

Input
6,000 tokens
a log excerpt or a finding
Output
800 tokens
a short analysis
Prices as of
2026-11-16

Security builds

Uncensored models and their prices per million tokens
Cyber Frontiercyber-frontierSecurity$5.25$15.75
Gemma 4 26B A4B Cybersecuritygemma-4-26b-a4b-it-cybersecuritySecurity$0.4224$1.3376
GLM-5.3 Flash Cybersecurityglm-5.3-flash-cybersecuritySecurity$0.5093$1.6975
OrcaCyber MiniMax M3.1 Flashminimax-m3.1-flash-cybersecuritySecurity$2.00$4.00
OrcaCyber Qwen3.8 Flash Nextqwen3.8-flash-next-cybersecuritySecurity$2.00$4.00
OrcaCyber Zero 1.5orca-zero-1.5-cybersecuritySecurity$7.50$18.75
Qwen3.8 27B Cybersecurityqwen3.8-27b-cybersecuritySecurity$0.40$2.40

General uncensored builds

Uncensored models and their prices per million tokens
GLM-5.3 Uncensoredglm-5.3-uncensoredUncensored$0.282$0.846
DeepSeek V4.1 Flash Abliterateddeepseek-v4.1-flash-abliteratedAbliterated$0.8488$1.6975
GLM-5.3 Abliteratedglm-5.3-abliteratedAbliterated$2.1825$4.365
DeepSeek V4 Pro Uncensoreddeepseek-v4-pro-uncensoredUncensored$0.97$1.94
Kimi K3 Uncensoredkimi-k3-uncensoredUncensored$5.745$28.68

Why these builds

The complaint in every security thread is the same: the stock frontier models decline the question, and the vendors gate security work behind verification programs. The builds on this page answer it. The security ids answer as an analyst would, with the identifiers, the evidence and the fix; the uncensored ids answer in whatever shape your prompt asks for. Each model page shows the daily check, and the key works in the agents below today. The service is for the assessment of systems the customer is authorized to test; the acceptable use policy has the line. Or send model auto/security and let the security profile of the auto router pick the red-team build for each request. Teams that want a site built only for this work can call the same builds on PentestLLM, which adds setup guides for coding agents, a CVSS calculator and a rules-of-engagement builder.

One request

curl
1curl https://api.unrestrictable.ai/v1/chat/completions \
2 -H "Authorization: Bearer sk-unr-live-…" \
3 -H "Content-Type: application/json" \
4 -d '{
5 "model": "cyber-frontier",
6 "messages": [{"role": "user", "content": "Write a Sigma rule that detects CVE-2024-3094 exploitation attempts and explain the detection logic."}]
7 }'

Agent tools

Each of these open-source agents takes an OpenAI-compatible endpoint. The settings are the ones their READMEs name; swap the model id for any row above.

shell
1# Strix (usestrix/strix): any OpenAI-compatible endpoint
2export STRIX_LLM="openai/cyber-frontier"
3export LLM_API_KEY="sk-unr-live-…"
4export LLM_API_BASE="https://api.unrestrictable.ai/v1"
5strix --target ./app
shell
1# HackingBuddyGPT (ipa-lab/hackingBuddyGPT)
2python -m hackingBuddyGPT.cli LinuxPrivesc \
3 --llm.api_key "sk-unr-live-…" \
4 --llm.api_url "https://api.unrestrictable.ai/v1" \
5 --llm.model "cyber-frontier"
env
1# CAI (aliasrobotics/cai): .env
2OPENAI_API_KEY="sk-unr-live-…"
3OPENAI_BASE_URL="https://api.unrestrictable.ai/v1"
4CAI_MODEL="cyber-frontier"

Daily check

ModelVerdictContext
Cyber FrontierNot checked yet1024K
Gemma 4 26B A4B CybersecuritySteady since Oct 29256K
GLM-5.3 Flash CybersecuritySteady since Oct 231024K
OrcaCyber MiniMax M3.1 FlashSteady since Nov 15256K
OrcaCyber Qwen3.8 Flash NextSteady since Nov 6256K
OrcaCyber Zero 1.5Changed Nov 14976K
Qwen3.8 27B CybersecuritySteady since Oct 30256K
GLM-5.3 UncensoredSteady since Oct 12not published
DeepSeek V4.1 Flash AbliteratedSteady since Oct 19not published
GLM-5.3 AbliteratedSteady since Oct 30not published
DeepSeek V4 Pro UncensoredSteady since Oct 12not published
Kimi K3 UncensoredSteady since Nov 3256K

Verdicts from the daily real-model check at render; the full list is on the proof page.

Refusals

The uncensored builds rarely refuse. If one does on a prompt that is within policy, send the request id to support; the daily check on the proof page shows which builds still answer the way they should.

Questions

Yes. 7 security builds and 5 general uncensored builds answer on one OpenAI-compatible key at https://api.unrestrictable.ai/v1, from $0.282 in and $0.846 out per million tokens on GLM-5.3 Uncensored. They answer CVE analysis, exploit triage, Sigma and YARA detection rules, malware analysis and reverse-engineering questions for systems you are authorized to test.

One that answers the security question and fits the budget. Cyber Frontier (cyber-frontier) leads here at $5.25 in and $15.75 out per million tokens, with a security-analyst instruction built into the route. GLM-5.3 Uncensored (glm-5.3-uncensored) is the budget pick at $0.282 in and $0.846 out. The agent tools above take any of them through the same base URL.

For a tool that reads long logs or a whole repository, the widest context window: Cyber Frontier at 1024K tokens. For a chat analyst that cites CVEs and writes the finding, the security builds. For a model you steer with your own system prompt, the general uncensored builds in the second table.

The 5 general uncensored builds here answer what a stock build of the same weights declines; each model page shows the daily real-model check. The security builds carry a security-analyst instruction ahead of your prompt, so the answer arrives as a finding with its evidence, its proof of concept and its fix.

A local 7B or 8B security model fits a laptop GPU and knows less than a 2026 frontier model. The builds here are full-size open-weight models served by API, so there is no GPU to buy and the id goes into the same tools a local model would. GLM-5.3 Uncensored is $0.282 per million input tokens.

Yes, and the open-source agents expect an OpenAI-compatible endpoint: Strix takes LLM_API_BASE, HackingBuddyGPT takes --llm.api_url, CAI takes OPENAI_BASE_URL. Set each to https://api.unrestrictable.ai/v1 with your key and a model id from the table.

Two different things share the phrase. Red-teaming an LLM means attacking a model or an AI app to find its failures (prompt injection, data leaks); tools like Promptfoo and DeepTeam do that. This page is the other direction: LLMs for red teamers, models that help a security engineer assess systems the customer is authorized to test.

The model answers; the operator decides what to run. Every request here is billed to a prepaid key, the daily check shows which model answered, and the service is for assessment of systems the customer is authorized to test, as the acceptable use policy says.

Join on Telegram