Z.ai

AI · Cybersecurity · China Tech

Launch

Z.ai's GLM-5.3 beats Anthropic's Mythos 5 on cybersecurity benchmark

The upgrade came purely from extended post-training on an unchanged base model, not a bigger one, yet investors still sent shares lower.

• Chinese AI firm Z.ai (Zhipu) launched GLM-5.3, claiming it scored 84.5% on the CyberGym cybersecurity benchmark, edging out Anthropic's Mythos 5 (83.8%) and OpenAI's GPT-5.6 Sol (83.6%).
• Z.ai, also known as Zhipu AI, is a Beijing-based, Hong Kong-listed AI developer whose GLM model line competes directly with OpenAI and Anthropic; it went public in January 2026 at a market cap near $7bn.
• GLM-5.3 keeps the same 753-billion-parameter mixture-of-experts base as June's GLM-5.2, with all performance gains coming from a scaled-up post-training process rather than a larger model.
• The model still trailed Mythos 5 on ExploitBench, a harder test of turning found flaws into working exploits, scoring 54.4% versus Mythos 5's 78%, showing offensive capability still lags detection.
• Working with Chinese security teams, Z.ai says GLM-5.3 surfaced 2,436 vulnerabilities across 269 real-world software projects, with 1,097 rated medium-to-high severity.
• Z.ai's Hong Kong-listed shares fell nearly 4% the day of the launch, with one analyst telling Bloomberg the company's commercial footing remains unsustainable as agentic AI inference costs rise.
• Cybersecurity had been a rare area where Chinese models lagged US frontier labs, so a benchmark win here signals the performance gap is narrowing even in domains built around sensitive, restricted capabilities.