Reassessment of the Capability and Cost Frontier: Maintain Overweight on Zhipu; MiniMax Still Awaits M3.1 Validation

Overview

The report assesses the commercial competitiveness of Chinese AI model vendors through a Pareto frontier of model capabilities and pricing. JPMorgan prefers Zhipu, which sits on the capability frontier and whose GLM-5.3 improvements are driven by its own iteration; it maintains a Neutral rating on MiniMax because its improvement benefits more from DeepSeek's price increase, while M3.1 and Hailuo H3 still need to demonstrate sustainable competitiveness and monetization potential.

Core views

Capability-leading models can address more complex, higher-value workloads and retain pricing power, while greater supply in mature-capability use cases will intensify price competition. Sustainable cost leadership must rest on structural advantages in model architecture, inference systems, and service efficiency, rather than simply aggressive pricing.

Analysis framework

The report applies a capability–cost Pareto framework: a model is on the frontier if competitors cannot offer higher capability at the same or lower price, or comparable capability at lower cost. Costs are primarily referenced to blended token pricing, while capabilities incorporate benchmark tests and actual product performance; the framework is used for relative positioning, not to precisely calculate customer return on investment.

Methodology notes

• Competitive Positioning FrameworkCapability–Cost Pareto Frontier

  Relative efficiency of model capabilities and pricing

  A model on the frontier has no direct alternative that is simultaneously more capable or cheaper; models inside the frontier face a more difficult commercialization environment.

• Valuation MethodologyDiscounted Forward Price-to-Earnings

  Target-price valuation based on expected 2030 earnings

  Target prices for both companies are based on 20x expected 2030 price-to-earnings and discounted using a 15% weighted average cost of capital.

Asset mapping & comparison

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

• Z AI Co Ltd - H（2513.HK）

  Core recommended name, rated Overweight

  Strengths

  GLM-5.3 improves coding and agent capabilities; the company has repeatedly launched domestic frontier models; capability improvements are internally driven, offering potential for high-value workflows and pricing power.

  Weaknesses

  Continued model iteration and broader independent validation are still needed; near-term profitability remains under pressure.

  Comparison

  Closer to the capability frontier than MiniMax, with more self-driven competitive improvement; it still faces competition from frontier models such as Kimi and DeepSeek.

  Risks

  Export controls, geopolitical and entity-list risks; intensifying competition; high R&D spending; uncertainty around commercialization and customer adoption; dependence on computing infrastructure and external suppliers.

• MiniMax Group Inc - H（0100.HK）

  Covered name, rated Neutral

  Strengths

  Offers a multimodal portfolio spanning text, audio, images, video, and productivity tools; serves both consumer and enterprise customers; internationalization and developer adoption provide long-term optionality.

  Weaknesses

  The current M3 has not established a clear edge in either capabilities or cost effectiveness; some improvement stems from competitors' price increases; independent vendors have weaker value capture than integrated platforms.

  Comparison

  At an earlier stage of catching up in model capabilities relative to Zhipu; faces DeepSeek's cost benchmark as well as the synergies ByteDance and Kuaishou enjoy across content, distribution, and advertising ecosystems.

  Risks

  Litigation with U.S. film studios; intensifying competition; high R&D spending and profitability pressure; uncertainty around commercialization and customer adoption; dependence on computing capacity and suppliers.

Key data

• Zhipu Rating and Target PriceOverweight; HK$1,800Raised from the previous HK$1,600; equivalent to approximately 20x expected 2027 US$5bn ARR.
• MiniMax Rating and Target PriceNeutral; HK$260Raised from the previous HK$160; equivalent to approximately 12x expected end-2026 US$1bn ARR.
• Zhipu Revenue Forecast RevisionRaised by 6%–10% for 2026–2030Reflects GLM-5.3 capability improvements and better commercialization expectations.
• MiniMax Revenue Forecast RevisionRaised by 11%–21% for 2027–2030Reflects more moderate model-pricing competition and Hailuo H3 multimodal optionality.
• Zhipu Current Share PriceHK$1,270Closing price on 2026-08-14.
• MiniMax Current Share PriceHK$329Closing price on 2026-08-14.

Impact & implications

The report shifts the industry's valuation focus from pure usage volumes or single model releases toward the ability to repeatedly sustain capability or cost advantages. If independently validated and translated into higher paid conversion, Zhipu's capability advantage could support its valuation; while MiniMax has long-term optionality from multimodality, internationalization, and dual consumer-enterprise product positioning, value capture and profitability remain key constraints for an independent vendor competing against integrated platforms such as ByteDance and Kuaishou.

Risks

• Model competition is accelerating, and frontier capabilities and pricing advantages may change rapidly.
• DeepSeek may adjust its pricing strategy, reducing MiniMax's short-term benefit.
• GLM-5.3's actual product performance and paid conversion may not meet expectations.
• MiniMax M3.1 and Hailuo H3 may fail to establish sustained leadership or effective monetization.
• High R&D spending, computing costs, and supply-chain constraints may compress profitability.
• Regulatory, export-control, geopolitical, and litigation risks may affect valuation and operations.

What to watch

• Independent validation of GLM-5.3's capabilities, user feedback, adoption rates, and paid conversion.
• Zhipu's continued iteration in post-training, data quality, reinforcement learning, and inference optimization.
• The capability, pricing, and cost-effectiveness positioning of MiniMax M3.1.
• Paid usage of Hailuo H3, pricing sustainability, and multimodal monetization.
• Changes in DeepSeek API pricing and the sustainability of its cost leadership.
• Competitive actions by integrated platforms across models, content creation, distribution, and advertising ecosystems.