Goldman Sachs: Knowledge Atlas Technology/Zhipu AI (2513.HK, Not Covered): Investor Meeting Highlights — Coding Strength and Path to Scaling

Goldman Sachs hosted an investor meeting for Zhipu AI on April 13–14, 2026, in Hong Kong, highlighting the company's competitive advantages in coding and agentic AI. Zhipu AI's GLM model ranks highly on the Berkeley Function Calling Leaderboard (BFCL), which evaluates an LLM's ability to accurately execute function calls. Demonstrating robust pricing power, the company has raised API prices by 100% year-to-date. While GLM 5.1 input prices remain significantly higher than domestic peers, they are still well below US competitors. Management noted that current demand outstrips compute supply by a multiple, with its coding models widely adopted by leading internet companies for complex programming tasks. To maximize hardware efficiency, Zhipu AI primarily leases compute power and servers from third parties, leveraging its proprietary AI infrastructure capabilities. Strategically, the company focuses on high-value production scenarios over lightweight use cases. Looking ahead, management views autonomous agents as the next major frontier. These agents are expected to feature self-learning capabilities, integrated control layers, and the ability to execute ultra-long-horizon tasks, potentially establishing a new scaling law. Additionally, vision-language model (VLM) multimodal understanding enables these models to interact with applications similarly to human users. The report includes a chart illustrating the ARR growth of Zhipu AI's open platform and API business alongside major model updates.