China Artificial Intelligence

This report examines the commercialization of open-weight LLMs among Chinese providers, highlighting that 'winner-takes-more' dynamics favor SOTA models that can bundle open-source distribution with high-quality proprietary API services and workflow tools.

Key Takeaways

• 1.Open-weight commercialization in LLMs is evolving into a 'winner-takes-more' framework where model capability and quality determine whether providers can monetize via proprietary API routes or if they face commoditization.
• 2.Zhipu AI's GLM-5.2 demonstrates a stronger monetization path via open-weights compared to MiniMax, justifying a target price increase for Zhipu while lowering MiniMax.
• 3.Proprietary model harnesses (e.g., coding agents, workflow tools) are becoming essential to convert open-weight adoption into sticky, first-party API revenue.

Table of Contents

• Equity Ratings and Price Targets
• Open-weight strategy: distribution upside, model-access pressure and a more flexible monetization playbook
• Open-weight is a dynamic commercial choice for LLM providers
• API quality: proprietary APIs sell stable access to fresher and higher-quality models
• Case Study #1: DeepSeek V4 Pro – official route wins through list price and cache economics
• Case Study #2: MiniMax M3 – similar headline price, better official-route quality
• CSP & inference platform distribution: margin retention versus model quality
• Workflows: proprietary harnesses can turn open-weight adoption into stickier API usage
• Company view: SOTA models should capture most of the open-weight monetization upside
• Zhipu AI — Maintain OW, raise PT to HK$2,000
• MiniMax – Maintain Neutral, lower PT to HK$300

Document Preview

Page 1 of 5

Subscribe for full accessStart Free Trial