QAX-GPT 奇安信安全大模型

What people say

There's no G2, Capterra or Gartner Peer Insights listing for QAX-GPT — it doesn't appear to be sold as a standalone product outside China's enterprise and government security market, so every performance figure below comes from Qi-Anxin's own published material, not third-party review sites.

By the company's account, the current version of the QAX-GPT Security Robot processes alerts at more than 60 times manual-analyst speed, with a false-positive rate roughly half that of human triage and a miss rate around 5% of the human baseline — Qi-Anxin describes its judgment quality as approaching an intermediate security analyst. Investigation and response time reportedly dropped from hours to minutes, with single-incident processing time cut by 98%. None of these figures have independent verification available in English-language sources.

What is independently checkable: QAX-GPT topped China's CyberSec-Eval benchmark, placing first overall and in all eight individual evaluation categories, and it passed the Cyberspace Administration of China's "dual filing" (双备案) review covering training-data security, model safety, content labeling and evaluation testing — a real regulatory gate for LLMs sold in China, not a vendor self-assessment. Qi-Anxin was among the first companies to receive a formal large-model security-capability certification when the product first shipped in August 2023, with the current generation launched industry-wide in March 2024.

The product ships as the QAX-GPT Security Robot and is also woven into Qi-Anxin's broader stack — Tianyan (threat intelligence), Tianqing (endpoint), Jiaotu, and its AISOC platform — rather than sold as a single bolt-on. There's no evidence of an English-language product page, developer API, or ASEAN go-to-market; procurement, language and data-sovereignty requirements would all need to be worked out case by case for any organization outside mainland China. For a China-based SOC already running Qi-Anxin's security stack, it's a credible alert-triage layer with real regulatory backing. For anyone evaluating it from outside that market, the absence of independent review data is the main limiting factor, not the technology itself.

Summary of public user & expert reviews, compiled by RECATOOLS.