GLM-5.3 Capability Improvements and DeepSeek Price Increase Lead Morgan Stanley to Remain Bullish on Z.AI

Overview

This report tracks recent developments in China's artificial intelligence foundation models, focusing on Z.AI's release of GLM-5.3 and the industry impact after DeepSeek formally launched the V4 series and raised prices. Morgan Stanley believes that competition is shifting from pure price competition toward model intelligence and commercialization quality, and remains bullish on Z.AI.

Core views

By significantly expanding post-training, GLM-5.3 achieved a major capability leap without changing the GLM-5.2 backbone, demonstrating that the existing backbone retains room for further improvement. Early feedback on DeepSeek V4 Pro has been relatively weak, while its price increase erodes its low-price advantage; the report believes this will help ease price competition in China's large-model market. For Z.AI, the release of globally leading models, international expansion through partnerships with overseas cloud service providers, and easing industry competition represent potential upside catalysts.

Analysis framework

The report assesses internal model-capability evaluations and early user feedback, peer-product pricing changes, and China's frontier large-model competitive landscape; it values Z.AI using a discounted cash flow method.

Methodology notes

• Valuation MethodDiscounted Cash Flow Method (DCF)

  Estimating enterprise value by discounting future cash flows

  The report values Z.AI using DCF, assuming a weighted average cost of capital of 15% and a terminal growth rate of 3%.

• Relative ValuationPrice-to-Sales Ratio (P/S)

  Revenue multiple implied by the target valuation

  The report states that its target price corresponds to 42x 2027 P/S.

• Technical EvaluationPost-Training Scaling

  Improving model capabilities through longer and richer reinforcement-learning post-training

  GLM-5.3's main capability gains stem from post-training scaling rather than a replacement of the base model.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• Z.AI CO., LTD.（2513.HK）

  Covered company and primary beneficiary

  Strengths

  GLM-5.3 has made progress in coding and cybersecurity capabilities; post-training scaling demonstrates model potential; it plans to release full weights; it has opportunities for partnerships with overseas cloud service providers and global expansion.

  Weaknesses

  Model competitiveness and commercialization delivery still require continued validation.

  Comparison

  Compared with DeepSeek V4 Pro, the early feedback cited in the report indicates that the latter is weaker in long-horizon tasks, coding capabilities, and stability, and no longer occupies the lowest-priced flagship-model position after its price increase.

  Risks

  Compute constraints, model performance lagging peers, and geopolitical risks.

• DeepSeek V4 Series

  Industry competitor and pricing bellwether

  Strengths

  Formally launched the V4 series.

  Weaknesses

  Early feedback on V4 Pro has been relatively negative regarding long-horizon tasks, coding capabilities, and stability; prices have increased significantly.

  Comparison

  After the price increase, V4 Pro is no longer China's cheapest flagship large model.

  Risks

  If performance and stability do not improve, the price increase may affect competitiveness and user adoption.

Key data

• GLM-5.3 Coding Capability Improvement50% improvement versus GLM-5.2Based on the internal evaluations described in the report.
• GLM-5.2/5.3 Base Model Parameter ScaleOver 700 billion parametersGLM-5.3 retains the GLM-5.2 base model.
• DeepSeek V4 Pro Price ChangeNon-cached input +1.25x; cached input +8x; output +2.38xBased on blended off-peak and peak-period pricing.
• DeepSeek V4 Flash Price ChangeNon-cached input +1.25x; cached input +2.75x; output +2.38xBased on blended off-peak and peak-period pricing.
• Z.AI Valuation Assumptions15% WACC; 3% terminal growth rateThe report uses DCF valuation.
• Valuation Implied by Z.AI Target Price42x 2027 P/STarget-price-implied valuation disclosed in the report.

Impact & implications

If GLM-5.3's capability improvements and open-weight plan gain market validation, Z.AI's technological competitiveness and appeal to developers may strengthen. DeepSeek's price increase and volatility in V4 Pro's early performance may reduce the market's reliance on ultra-low-priced flagship models and shift the industry's focus toward performance, stability, and unit economics; however, this assessment remains dependent on subsequent product testing, customer adoption, and competitor iteration.

Risks

• Constraints on computing resources may limit model training, inference, and commercialization expansion.
• If Z.AI's model performance lags peers, expectations for technological leadership and valuation may come under pressure.
• Geopolitical risks may affect supply chains, access to computing power, and global business expansion.
• Large-model products iterate rapidly, and early internal evaluations and user feedback may not represent the long-term competitive landscape.
• Morgan Stanley discloses that it has or seeks investment-banking service relationships with Z.AI and certain related companies; investors should consider potential conflicts of interest.

What to watch

• Progress toward GLM-5.3's planned open-weight release following security hardening, and developer feedback.
• Third-party testing and customer adoption of GLM-5.3 in coding, cybersecurity, and complex tasks.
• Subsequent improvements to DeepSeek V4 Pro in stability, long-horizon tasks, and coding capabilities.
• Changes in pricing, demand, and the competitive landscape of China's large-model market following DeepSeek's price increase.
• Progress in Z.AI's partnerships with overseas cloud service providers and global expansion.