Skip to content
AI for penetration testing · 9 security builds

Pentest AI models. One API key.

Security-tuned and abliterated models for authorized penetration testing, red teaming, bug bounty and CTF work, in the agent you already run. Prepaid from $10 by card; no subscription.

Get a keySee a request
POSThttps://api.pentestllm.ai/v1/chat/completions

9 security builds and prices · 5 of 5 passed today’s check · agent setup guides · authorized testing only

free turn, no sign-up
OpenCodeClaude CodeCursorClineAiderKali MCPPentestGPTOpenAI SDK

The security catalog. Cybersecurity-tuned and abliterated builds, priced per token, as the catalog stood on 2026-11-11.

Rows of server cabinets in a data centre

Security builds

9

Cybersecurity-tuned open-weight builds on one prepaid key, in the OpenAI and Anthropic wire formats.

Output price per million tokens, lowest first

$1.3376 to $18.75

5/5

Passed today’s model check

1M

Largest context window; 7 builds take tool calls

One key. The router, the tools and the receipt come with it.

Each request goes to the build’s live route and is billed for the tokens it used. Nothing in a request body is stored.

See a request
{ "model": "auto/security", "messages": [ … ] }
fastGemma 4 26B A4B Cybersecurity, GLM-5.3 Flash Cybersecurity
balancedRed Team (Flash), OrcaCyber MiniMax M3.1 Flash
frontierRed Team (Pro), Cyber Frontier

auto/security picks the build

Send one model id. A fast tier answers short questions, a balanced tier the usual work, a frontier tier the hard ones. You pay the price of the build that answered.

Input, 40,000 tokens at $0.40 per million
$0.02
Output, 1,200 tokens at $2.40 per million
$0.0029
This turn
$0.02

Billed to the token

One agent turn of 40,000 tokens in and 1,200 out on Qwen3.8 27B Cybersecurity costs $0.02. A $10 top-up covers about 529 of them.

Web search$0.01 per search
Web fetch$0.001 per page fetched
Code sandbox$0.05 per container-hour, 1-minute minimum
Remote MCPNo tool fee

Tools that run on our side

Declare web search, page fetch, a code sandbox or a remote MCP server in the request. The gateway runs the tool and bills its fee with the tokens.

Gemma 4 26B A4B Cybersecuritysteady since Oct 29
OrcaCyber Qwen3.8 Flash Nextsteady since Nov 8
OrcaCyber Zero 1.5steady since Nov 7
Qwen3.8 27B Cybersecuritysteady since Oct 29

A daily check that each build is the build

Every day each model answers a fixed test and the result is published. Today 5 of 5 checked security builds are steady.

9 security builds, live prices. USD per million tokens, read from the catalog on 2026-11-11.

All modelsPrice index

From key to first run. Three steps.

  1. Add $10 and create a key.

    By card, no subscription. Keys start with sk-ptl-.

  2. Point your agent at it.

    Pick the tool and the build on the right and copy the file. Each agent also has a full guide.

  3. Run it against what is in scope.

    Systems you own, or hold written permission to test. The rules of engagement builder writes the letter.

Add $10Agent guidesScope letter builder
Patch cables in a network switch
opencode.jsonc
{
  "$schema": "https://opencode.ai/config.json",
  "model": "pentestllm/qwen3.8-27b-cybersecurity",
  "providers": {
    "pentestllm": {
      "name": "PentestLLM",
      "env": ["PENTESTLLM_API_KEY"],
      "package": "@opencode/ai/providers/openai-compatible",
      "settings": {
        "baseURL": "https://api.pentestllm.ai/v1"
      },
      "models": {
        "qwen3.8-27b-cybersecurity": { "name": "Qwen3.8 27B Cybersecurity" }
      }
    }
  }
}

Built for the work. Authorized testing only: systems you own, or hold written permission to test.

Rows of server cabinets in a data centreRed teamModels for authorized adversary emulation: planning, tooling review, detection notes and reporting.Patch cables in a network switchBug bountyModels for scope reading, recon notes, code review and report drafting on programs you are enrolled in.Close view of a circuit boardCTFModels that reason through web, crypto, reversing, forensics and pwn challenge categories.Components on a printed circuit boardMalware analysisModels for static triage, behaviour summaries and detection rules. Analysis and detection only.

Compare, and go deeper

What each alternative is, what it costs today, and where a per-token endpoint fits. Plus the long questions people ask about security models.

Components on a printed circuit board

Prepaid, per token. Top up once, spend it at each build’s listed price. When it runs out, requests stop. Full pricing

Start

Try every build at the listed price.

One payment. Nothing renews.

9 security builds, The auto/security router, Pay per token, No subscription

$10top-up

Add $10

Engagement

For a tester with work on the calendar.

One payment. Nothing renews.

Everything in Start, +5% bonus credit, Spend limits per key, Signed monthly statements +5% credit

$200top-up

Add $200

Team

For a red team or a consultancy.

One payment. Nothing renews.

Everything in Engagement, +15% bonus credit, Saved card and auto top-up, Several members, one balance +15% credit

$1,000top-up

Add $1,000
FAQs

Common questions

Short answers, with the live numbers where a number is the answer.

Talk to a human

Read the docs

Or email hello@pentestllm.ai

An API for security-tuned and abliterated language models, sold prepaid and priced per token. One key reaches 9 security builds from OpenAI-format and Anthropic-format clients.

People doing authorized security work: penetration testers, red teams, bug bounty hunters, CTF players and analysts. You must own the systems you test or hold written permission to test them. The acceptable use policy at /acceptable-use says what is not allowed.

Far less often than a general assistant. The builds are tuned or abliterated for security work, so they answer questions about vulnerabilities, tooling and detection that other models decline. That is reduced refusals, not zero. Today 5 of 5 checked builds answered the daily test as the model they are named after.

You pay per token at each build's listed price, from a prepaid balance. Qwen3.8 27B Cybersecurity is $0.40 in and $2.40 out per million tokens, the lowest input price here as the catalog stood on 2026-11-11. The smallest top-up is $10, by card, with no subscription. Larger top-ups add bonus credit: 5% at $200, 10% at $500, 15% at $1,000.

Yes. Any client that accepts an OpenAI-compatible base URL works with the key, and the API also takes Anthropic-format requests on /v1/messages; 7 of the 9 security builds take tool calls. The agent guides at /agents give the exact config for each tool.

A router id. Send auto/security as the model and the gateway picks a security build by how hard the request is: a fast tier for short questions, a balanced tier, and a frontier tier for long or difficult work. You are billed at the price of the build that answered.

No. Prompts and completions pass through and are not written anywhere, unless you turn on replay for a key, which keeps that key’s request bodies for 7 days. The log keeps the time, the model, token counts, the cost and a SHA-256 fingerprint.

Put a security model in your agent

One key, 9 security builds, prepaid from $10. For systems you are authorized to test.

Get a key Read the docs
  • OpenAI and Anthropic wire formats
  • The auto/security router
  • Web search, fetch and a code sandbox
  • No request bodies stored
  • No subscription