Is PentestLLM a replacement for PentestGPT?

Not directly. PentestGPT is an agent and PentestLLM is a model API. You can run PentestGPT’s interactive mode on this endpoint, or use a different agent such as OpenCode with the same key.

The software is open source under the MIT licence, as its README stated in November 2026. It needs a language model to work, and that model is billed by whichever provider or subscription you connect.

Which models does PentestGPT support?

In November 2026 its README says the autonomous pipeline drives Claude Code or Codex, and the legacy interactive mode supports OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Moonshot and local Ollama.

Why would I change the model behind an agent?

General models decline many offensive security questions, and one declined step can stall an agent run. Security-tuned builds answer those questions far more often, which is reduced refusals, not zero.

What does PentestLLM cost?

Per token from a prepaid balance with a $10 minimum top-up and no subscription. On 2026-11-11 input prices for security builds start at $0.40 per million tokens.

A PentestGPT alternative by API

PentestGPT is an open-source penetration testing agent that needs a language model behind it; PentestLLM is a model API, so it is less a replacement than the endpoint you run PentestGPT or another agent on. One prepaid key from $10 reaches 9 security-tuned builds that answer the security questions general models decline.

Get a key Prices per token

shell

curl https://api.pentestllm.ai/v1/chat/completions \
  -H "Authorization: Bearer sk-ptl-live-…" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "auto/security",
    "messages": [{"role": "user", "content": "Rank these scanner findings by severity and say what to verify first."}]
  }'

What PentestGPT is, as of November 2026

PentestGPT is a research project published at USENIX Security 2024 and released under the MIT licence. Its README, read on 11 November 2026, describes two modes. The newer one is an autonomous pipeline that drives the Claude Code or Codex command-line tools through staged phases. The older one, kept as pentestgpt-legacy, is the interactive assistant from the paper: three cooperating model sessions that maintain a task tree while you run the commands.

The legacy mode lists OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Moonshot and local Ollama as providers, each configured with its own API key. PentestGPT itself is free. The model behind it is not: you bring a subscription login or a provider key, and the provider’s refusal behaviour comes with it.

Why people look for an alternative

Refusals: An agent is only as willing as its model. On a general model, a declined step can stall a run partway through.
Cost shape: A chat subscription has a cap that an agent can exhaust in an afternoon. Per-token billing scales with the work.
Control: Some testers want to choose the model per task and see what each request cost.
Other agents: Some simply prefer OpenCode, Cline or aider and want a security model behind those.

What PentestLLM adds

PentestLLM does not ship an agent. It serves models: 9 security-tuned builds and the abliterated builds beside them, behind an OpenAI-compatible endpoint at https://api.pentestllm.ai/v1 and an Anthropic-compatible Messages endpoint. The id auto/security routes each request to a fast, balanced or frontier build and bills it at that build’s price. Web search, page fetch, a code sandbox and remote MCP run on the API side when a request declares them.

The builds are tuned or abliterated to answer security questions other models decline. That is reduced refusals, not zero, and it is for authorized testing only.

Running PentestGPT on this endpoint

PentestGPT’s legacy mode reads an optional OPENAI_BASE_URL override and takes a --base-url flag, so its OpenAI connector can be pointed at https://api.pentestllm.ai/v1. Its model list is a fixed registry, so there is one extra step to make it send an id this API serves. The PentestGPT setup guide walks through it and says plainly which parts of PentestGPT we could not verify against this endpoint.

Or use a different agent

If what you want is an assistant in a terminal that plans, runs tools with your approval and keeps notes, a general coding agent on a security model does that today. OpenCode takes a provider block in one config file. Claude Code takes two environment variables. For tool execution on a Kali machine, the Kali MCP guide connects the Kali project’s own MCP package.

When PentestGPT is the better choice

If you want the staged pipeline and task tree from the paper, use PentestGPT; nothing here replaces that design. If you already pay for Claude Code or Codex and your work stays inside what those models accept, its autonomous mode uses those logins directly. PentestLLM is for the cases where the model is the limiting part. Either way the same rule applies: test only systems you own or have written permission to test, as the acceptable use policy says.

9 security builds, priced per token

Read from the live catalog on 2026-11-11. Prices are US dollars per million tokens, paid from a prepaid balance that starts at $10. The id auto/security has no price of its own: each request is billed at the build that answered.

Build	Input	Output	Context
Qwen3.8 27B Cybersecurity	$0.40	$2.40	262,144
Gemma 4 26B A4B Cybersecurity	$0.4224	$1.3376	262,144
GLM-5.3 Flash Cybersecurity	$0.5093	$1.6975	1,048,576
Red Team (Flash)	$0.90	$3.00	not published
Red Team (Pro)	$1.50	$6.00	not published
OrcaCyber MiniMax M3.1 Flash	$2.00	$4.00	261,632
OrcaCyber Qwen3.8 Flash Next	$2.00	$4.00	261,632
Cyber Frontier	$5.25	$15.75	1,048,576
OrcaCyber Zero 1.5	$7.50	$18.75	999,488

Full price list and tool fees · Every model · API docs

Questions

Not directly. PentestGPT is an agent and PentestLLM is a model API. You can run PentestGPT’s interactive mode on this endpoint, or use a different agent such as OpenCode with the same key.
The software is open source under the MIT licence, as its README stated in November 2026. It needs a language model to work, and that model is billed by whichever provider or subscription you connect.
In November 2026 its README says the autonomous pipeline drives Claude Code or Codex, and the legacy interactive mode supports OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Moonshot and local Ollama.
General models decline many offensive security questions, and one declined step can stall an agent run. Security-tuned builds answer those questions far more often, which is reduced refusals, not zero.
Per token from a prepaid balance with a $10 minimum top-up and no subscription. On 2026-11-11 input prices for security builds start at $0.40 per million tokens.

One key, from $10

PentestLLM is for systems you own or have written permission to test. The acceptable use policy says what is not allowed.

Get a key

usage, 1,000 agent turns on Qwen3.8 27B Cybersecurity

Input, 8,000 tokens at $0.40 per million: $0.0032
Output, 1,000 tokens at $2.40 per million: $0.0024
One turn: $0.0056
1,000 turns: $5.60

output price per million tokens, 9 builds

$1.3376 Gemma 4 26B A4B CybersecurityOrcaCyber Zero 1.5 $18.75