Zhipu AI

Chinese AI lab developing GLM large language models and agentic AI systems.

Updated April 2026

Overview

Website: zhipuai.cn
Founded: 2019
Headquarters: Beijing, China
Segment: Frontier Foundation Model Labs
Posture: Regional / Emerging

Product overview

Zhipu AI builds the GLM series of multimodal large language models, including GLM-5 which rivals Claude Opus in coding and agent benchmarks, alongside products like ChatGLM chatbot, AutoGLM agents, and text-to-video model Ying. Primarily used by over 400,000 Chinese enterprises, government offices, and developers for business applications, customer support, and research. Distinct for training on domestic chips like Huawei Ascend amid U.S. sanctions, open-sourcing select models, and strong agentic capabilities optimized for Chinese market compliance.

Revenue model

Primarily API usage-based pricing per token (GLM-5.1: $0.54-$1.40/M input tokens, $4.40/M output); GLM Coding Plan subscriptions ($3/mo Lite, $9-15/mo Pro, custom Enterprise); enterprise licensing and MaaS for model fine-tuning. Commercial revenue doubled recently.

Moat

Zhipu AI's key competitive moat is its proprietary GLM series of large language models, particularly GLM-5 and GLM-4 variants, which rank among the global top 5-20 on benchmarks like Artificial Analysis and LMSYS Arena, excelling in Chinese-language tasks (SuperCLUE), hallucination suppression via self-developed multi-head attention mechanisms, coding (rivaling or outperforming GPT-4 and Claude Opus), and agentic capabilities for multi-step tasks. This technological edge, stemming from independent R&D on domestic chips like Huawei's, creates high switching costs for its 150,000+ paying users across 184 countries who favor its cost-effectiveness (dozens of times more token output per cost) and have driven subscription sell-outs, reinforced by open-source distribution for broad adoption and on-premise/cloud deployment flexibility in sectors like finance and healthcare.