Key Takeaways:
CMBI reiterated Buy on Z.AI (02513.HK) with a HKD1,503.9 target after GLM-5.3 launch lifted coding 50%.
GLM-5.3 significantly improved coding and Agent capabilities while striking a better balance between model performance and token efficiency, CMBI said in a research note. The broker flagged the model's preliminary cybersecurity strengths, particularly in the early stages of task chains, while saying there is room for improvement in full vulnerability exploitation.
The model, built on the same 753-billion-parameter mixture-of-experts base as GLM-5.2, scored 28.3 on Terminal Bench 3.0, up from 4.6, and 66.9 on DeepSWE v1.1, up from 46.2. On Agents' Last Exam, which tests cross-tool collaboration, it rose to 28.5 from 23.8. On CyberGym, which tests vulnerability discovery, it reached 84.5 percent, ahead of Anthropic's Claude Mythos 5 at 83.8 percent and OpenAI's GPT-5.6 Sol at 83.6 percent. On ExploitBench, which tests converting discovered flaws into working attacks, GLM-5.3 scored 54.4 percent, more than double GLM-5.2's 24.4 percent but below Claude Mythos 5's 78.0 percent. Z.ai said GLM-5.3 has found more than 2,400 vulnerabilities across 269 software projects, about half rated medium severity or higher.
Token efficiency also improved. At maximum reasoning effort, GLM-5.3 reached 34.5 percent accuracy on Z.ai's internal coding benchmark using about 75,000 tokens, versus 23.4 percent for GLM-5.2 at 96,000 tokens. The model's gains came entirely from post-training scaling, with the base model unchanged from GLM-5.2, according to Z.ai.
The stock fell 5.3 percent to HKD67.00 on Aug. 17, with short selling of HKD455.12 million at a 5.6 percent ratio. CMBI's target implies more than 2,000 percent upside from current levels. JPMorgan also raised its target on Z.AI and lifted revenue forecasts, pointing to growing institutional confidence in the company's growth trajectory.
Z.ai plans to release GLM-5.3's weights on Hugging Face under an open-source license within two weeks, following safety evaluations. The model is currently available through the company's GLM Coding Plan subscription service. CMBI said the launch will support the share price and remains optimistic on Z.AI's long-term growth prospects. The open-source release will test whether Z.ai can match the developer adoption of rivals such as DeepSeek and Alibaba's Qwen.
This article is for informational purposes only and does not constitute investment advice.
