China's GLM-5.2 model claims to match an Anthropic model on detecting security vulnerabilities — but the Anthropic model named in the comparison doesn't appear in any public lineup.
The Wall Street Journal reported that GLM-5.2, a Chinese AI model, performs on par with an Anthropic model called "Claude Mythos" on cybersecurity tasks, specifically identifying software flaws. The problem: Anthropic has no publicly listed product called "Claude Mythos." The company's known lineup runs through the Claude 3 family and the newer Claude 4 series. Neither the original report nor the source material explains whether "Claude Mythos" is a misprint, an unreleased internal model, a codename, or something that was simply made up.
Strip out the unverifiable model name and there's still a real story underneath: Chinese AI labs narrowing the gap with Western frontier models is a trend with concrete stakes. Security-specific AI — catching zero-days, scanning code for exploitable flaws before attackers do — is among the most strategically sensitive applications in the field. If GLM-5.2 genuinely competes at that level, the benchmark target matters less than the capability claim itself.
A comparison that hinges on a model nobody can locate in the product catalog is, at best, unfinished reporting — and at worst, a free publicity ride for a claim that can't be checked.