ARTEX vs PentestGPT:a platform with a console against a command-line agent.
PentestGPT began as a research project in 2023 and has 16,443 GitHub stars. ARTEX is a 2026 platform built as a Go service. This page compares them on design, license, model support, status and the numbers each project publishes, as of 2026-11-16.
Data as of 2026-11-16
ARTEX and PentestGPT side by side
| Aspect | ARTEX | PentestGPT |
|---|---|---|
| Type | A server: one Go binary with a web console over PostgreSQL | A command-line tool in Python, run as pentestgpt --target |
| License | AGPL-3.0 | MIT |
| Open or closed | Open code. The original repo was deleted on 2026-10-08 and forks carry on | Open. GreyDGL/PentestGPT is public on GitHub |
| Origin | Autumn-27, first commit on 2026-07-26 | GreyDGL, repo created 2023-02-27, with a USENIX Security 2024 paper |
| Architecture | A planner and 3 workers by default over an asset graph and an exploration graph, with a guard and a recording proxy | v1.0: a staged pipeline that drives Claude Code or Codex. The older interactive mode runs reasoning, generation and parsing sessions around a task tree |
| Model support | Any Anthropic or OpenAI compatible endpoint, with a failover chain per task | Autonomous mode: Claude Code or Codex backends. Interactive mode: OpenAI, Anthropic, Gemini, DeepSeek, xAI, Qwen, Moonshot and Ollama |
| Human control | Rules, then an LLM judge, then human approval. People can chat, add hints or pause a worker | The v1.0 agent runs without human intervention. Sessions can be saved and resumed |
| Status | RuoJi6/ARTEX released v0.3.30 on 2026-11-14 | Latest release v1.0.0 on 2025-12-24. Last push 2026-07-14. 16,443 stars |
| Public benchmark | Baidu Agent+ final: 1st overall (2026-09-03). Live range 67.83/100, 7th. TSecBench 89.78/100, 6th | 86.5% (90 of 104) on an XBOW validation suite, December 2025, called a historical result by its README |
Two different shapes of tool
PentestGPT is a command-line program. You install it with uv, point it at a target and it works through a staged pipeline: recon, exploit and walkthrough for a capture-the-flag challenge, or asset discovery, vulnerability identification and report for a pentest. Its v1.0 release moved to an agentic design that drives the Claude Code or Codex command-line tools to run the actual commands.
ARTEX is a service. One Go binary serves a web console and an API, keeps tasks, findings and two graphs in PostgreSQL, and runs a planner that hands work to several workers. Every tool call passes a guard of intercept rules, an LLM judge and human approval, and a recording proxy keeps the traffic as evidence. People work in the console, not in a terminal session.
Models and setup
PentestGPT's README says its autonomous pipeline can use Claude Code or Codex as the backend. A separate legacy mode, kept from the 2024 paper, supports many providers, including OpenAI, Anthropic, Google Gemini, DeepSeek, xAI, Qwen, Moonshot and local Ollama models. That mode runs three cooperating model sessions that keep a pentesting task tree.
ARTEX connects to any endpoint that speaks the Anthropic Messages, OpenAI Chat Completions or OpenAI Responses format, and each task can hold an ordered chain of model profiles that fails over on quota errors. Setup needs PostgreSQL, so it is heavier than a single command-line install. In return it keeps history, assets and findings across tasks.
Benchmarks are not comparable
The two projects publish different kinds of numbers. PentestGPT's README reports 86.5%, or 90 of 104 challenges, on an XBOW validation suite in December 2025, and it says this is a historical research result, not a guarantee for the current agent. It also says the supported command-line tool no longer includes an XBOW runner.
ARTEX's numbers come from a contest. It was first overall at the Baidu BSRC Agent+ final, which was judged half on a live range and half by a jury, and its live range score was 67.83 out of 100. It scored 89.78 out of 100 on the Tencent TSecBench blind range. A suite of web challenges and a live contest range test different things, so do not read one score against the other.
Status and license
PentestGPT is MIT licensed, has 16,443 stars and 2,879 forks, and its last push was on 2026-07-14, with the v1.0.0 release dated 2025-12-24. ARTEX is AGPL-3.0, so a network service built from it must publish its source. Its original repo is gone, so the active line is in forks such as RuoJi6/ARTEX at v0.3.30.
More comparisons
CompareAntiFox vs ARTEX
AntiFox is a public fork that says it was built on ARTEX and adds report delivery. What changed, how active it is and its license.
ARTEX vs Strix
A self-hosted Go platform against an Apache-2.0 Python agent with a Docker sandbox, a cloud edition and CI support.
ARTEX vs CAI
An AGPL platform still maintained in forks against CAI from Alias Robotics, now archived with a commercial successor.
ARTEX vs HexStrike AI
A self-contained agent platform against an MCP server that lets your own AI client call more than 150 security tools.
Follow the ARTEX story.
New forks, releases and reports, dated and sourced.
