GLM-5.3 unveiled with strong capabilities

CMB International Global Markets | Equity Research | Company FlashNote Z.AI (2513 HK) GLM-5.3 unveiledwith strong capabilities FLASH Z.ai unveiled (14 Aug) itsGLM-5.3(link).While the base model remainsunchanged from GLM-5.2, Z.ai has materially raised the model’s intelligenceceilingthrough aggressive post-training scaling.In our view,GLM-5.3demonstrates significant improvements in coding and agentic capabilities,whileachieving a better balance between performance and token efficiency.The model also exhibits emerging cybersecurity capabilities, with particularstrength inearlier stages of cybersecurity task-chain such as vulnerabilityvalidation, while leaving room for further improvement in full-chain exploitationtasks. We view the launch of GLM-5.3 as a positive catalyst for the stock andremain constructive on Z.ai’s long-term growth prospects. Reiterate BUY witha TP of HK$1,503.9. China Software & IT Services Saiyi HE, CFA(852) 3916 1739hesaiyi@cmbi.com.hk Ye TAO, CFA(852) 3850 5226franktao@cmbi.com.hk GLM-5.3 demonstrates significant improvements in coding andagentic capabilities.GLM-5.3 ranked No. 1 among open-source modelson public benchmarks including Terminal-Bench 3.0 and Agents’ LastExam (see Fig. 1). On Terminal-Bench 3.0, which measures a model’sability to complete complex tasks in real terminal environments, GLM-5.3’sscore increased to 28.3 from 4.6 for GLM-5.2. On DeepSWE v1.1, whichfocuseson long-horizon software engineering and sustained codemodification, its score rose to 66.9 from 46.2. On Agents’ Last Exam, whichcovers a broad range of real-world professional scenarios and emphasizescross-tool collaboration and long-horizon task execution,the scoreimproved to 28.5 from 23.8. In addition, GLM-5.3 achieved a betterbalance between model performance and tokenefficiency. Wentao LU, CFAluwentao@cmbi.com.hk Shuyin GUO(852) 3916 3716guoshuyin@cmbi.com.hk GLM-5.3 exhibits emerging cybersecurity capabilities.From a task-chain perspective, cybersecurity capabilities can broadly be divided intocode review, vulnerability validation, vulnerability exploitation, and closed-loop attack-and-defense execution in real-world environments. At present,GLM-5.3’s strengths are concentrated primarily in the earlier stages of thischain (see Fig. 2). Across the three evaluations, GLM-5.3 showed greaterimprovementover GLM-5.2 in tasks closer to the front end of thevulnerability-exploitation chain.As tasks progressed toward full-chainexploitation, the remaining gap versus frontier closed-source models suchas Mythos 5 widened, suggesting room for further capability improvement,in our view. Remainconstructive on Z.ai’s long-term growth prospects.Ourpositive view on Z.ai is anchored by its: 1) frontier model capabilities,underpinnedby deep research expertise that supports rapid modeliteration;2)strategic foresight in identifying monetization scenarios,positioning it well to capture opportunities yet to unfold; and 3) largeinstitutional customer base served with tailor-made solutions, alongsidestrong growth prospects for its open platform and API business.(seelinktoour initiation report:Sailingtoward the vast frontier of AGI, published on11 Jun 2026). Valuation and risks.Our target price of HK$1,503.9 is based on 40.0x2028E P/S, discounted back to 2026E at a 15% WACC.Key catalysts:1)better-than-expectedARRgrowth;and2)higher-than-expectedpenetration and monetization across new use cases and verticals.Keyrisks:1) slower-than-expected iteration of frontier models; and 2) moreintense industry competition than we expect. Source:Company data,CMBIGM Source:Company data,CMBIGM Disclosures& Disclaimers Analyst CertificationThe research analyst who is primary responsible for the content of this research report, in wholeor in part, certifies that with respect to the securities or issuer that the analyst covered in this report: (1) all of the views expressed accurately reflect his or her personal views about the subject securities or issuer; and (2)no part of his or her compensation was, is, or will be, directly or indirectly, related to the specific views expressed by that analyst in this report.Besides, the analyst confirms that neither the analyst nor his/her associates (as defined in the code of conduct issued by TheHong Kong Securities and Futures Commission) (1) have dealt in or traded in the stock(s) covered in this research report within 30 calendar days prior to thedate of issue of this report; (2) willdeal in or trade in the stock(s) covered in this research report 3 business days after the date of issue of this report; (3) serve as an officer of any of the HongKong listed companies covered in this report; and (4) have any financial interests in the Hong Kong listed companies coveredin this report. CMBIGM RatingsBUY : Stock with potential return of over 15% over next 12 monthsHOLD: Stock with potential return of +15% to-10% over next 12 monthsSELL: Stock with potential loss of over 10% over next 12 mo