Zhipu AI

What Zhipu AI does

Zhipu AI (legally Beijing Knowledge Atlas Technology Joint Stock Company Limited, trading as Z.ai; HKEX: 02513) is a Beijing-based AI company founded in 2019 that develops the GLM (General Language Model) family of foundation models and an integrated MaaS (Model-as-a-Service) platform. Its flagship GLM-5.2 is a 753-billion-parameter Mixture-of-Experts foundation model with a 1M-token lossless context window, released under an MIT open-source license with no regional restrictions. The company serves enterprise developers, Chinese hyperscalers and telecom carriers, government and SOE buyers, and global open-source adopters through a portfolio that includes GLM-5.2, GLM-5V-Turbo (multimodal vision coding), GLM-5-Turbo (agent-optimized), AutoGLM/AutoClaw autonomous agents, and consumer-facing products such as Z.ai, ChatGLM, Zlearn, Zread.ai, and AMiner.

Zhipu's core technology stack centers on its proprietary GLM-5.2 architecture with IndexShare-style efficient sparse attention, multimodal fusion for visual coding, and end-to-end training executed entirely on Huawei Ascend accelerators with no Nvidia hardware — a structural choice aligned with China's domestic compute sovereignty goals. Model outputs are exposed commercially through the Bigmodel MaaS platform, which offers multimodal APIs (GLM-5.2 priced at $1.40/$4.40 per million input/output tokens), enterprise-ready agent APIs (General Translation Agent, GLM Slide/Poster Agent), model fine-tuning services, and AI web-search tools. Distribution combines self-serve API and open-source self-hosting (HuggingFace, ModelScope) for developers with channel partnerships through Huawei Cloud's AI Model Partner Program, China Telecom/Mobile/Unicom token-priced consumer bundles, Alibaba Cloud's Bailian MaaS marketplace, and Feishu's (Lark) OpenClaw plugin.

Zhipu monetizes through usage-based API billing, the GLM Coding Plan subscription, freemium onboarding via a 20M-token free trial, and B2B2C token reselling through telecom carriers. FY2025 revenue reached 724.3 million yuan (~$100-107 million), up 131.9% year-over-year, against a net loss of 4.72 billion yuan amid heavy compute spend and intensifying competition from DeepSeek-V3.2 and other Chinese AI labs. The company completed a Hong Kong IPO in January 2026 at HK$116.2 per share (raising ~HK$4.35 billion / ~$558 million), saw its market cap cross HK$1 trillion by June 2026, and is pursuing a secondary Shanghai STAR Market A-share listing alongside a planned $2.2 billion follow-on offering.

Name

Zhipu AI

Legal name

Beijing Knowledge Atlas Technology Joint Stock Company Limited

Website

https://zhipuai.cn

Company type

Public

Founded year

2019

Operating status

Ipo

Headcount range

501–1,000 employees

Short description

Zhipu AI (trading as Z.ai, HKEX: 02513) is a Beijing-based AI developer that builds the GLM family of open-source foundation models and a MaaS API platform, serving enterprise developers, Chinese telecom and hyperscaler channels, government/SOE buyers, and global open-source adopters.

Ownership category

akta.pro rank

Product category

Foundation Models

NAICS

Custom Computer Programming Services (541511), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (5182)

SIC

Services-Prepackaged Software (7372), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM), Model Hosting, Serving & Inference Platforms (HDAAACAB), Enterprise Foundation Model Integration & APIs (Connectors, Governance, Deployment) (HDAAACAO)

Keywords

• Large language models
• Foundation model APIs
• Autonomous AI agents
• Multimodal AI models
• AI coding assistants

Headquarters

HQ city

Haidian

HQ country

China

HQ region

Asia

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel, Marketing or Sales, Operations

Revenue model

1. API / MaaS usage-based billing: Consumption-based revenue from the Bigmodel platform: developers and enterprises pay per token for GLM-5.2, GLM-5V-Turbo, GLM-5-Turbo, AutoGLM and dozens of vertical agents, with published price points of $1.40 / $4.40 per million input/output tokens for GLM-5.2.
2. GLM Coding Plan subscription: Recurring subscription product (the "GLM Coding Plan") sold alongside raw API access; demand has been chronically short, contributing to recent price increases.
3. Freemium onboarding (20M free tokens): Self-serve free trial of 20 million tokens via z.ai that funnels developers into paid API tiers; serves as a top-of-funnel land motion for the MaaS business.
4. Telecom-channel token reselling: Wholesale API access resold through China Telecom's token-priced AI packages (consumer plans from 9.9 yuan/month, business plans 39.9–299.9 yuan/month), bundled alongside TeleChat and DeepSeek-V3.2; effectively a B2B2C distribution revenue stream.
5. Enterprise / cloud-platform licensing: Model access licensed through hyperscaler MaaS marketplaces (Alibaba Cloud Bailian offers switchable access to Zhipu models at promotional 7.9 yuan lite / 39.9 yuan pro first-month rates), embedding Zhipu into enterprise procurement workflows.

Pricing tiers

Go-to-market motion6 records

Distribution channels8 records

Marketing channels8 records

Core offering

Zhipu AI develops and commercializes the GLM family of foundation large language models — including the flagship GLM-5.2 (753B-parameter MoE with 1M-token context), GLM-5V-Turbo (multimodal vision coding), GLM-5-Turbo (agent-optimized) and AutoGLM (autonomous agent). Models are delivered via the Bigmodel Model-as-a-Service platform with usage-based APIs, released as open-source weights on HuggingFace/ModelScope, and packaged into consumer products such as Z.ai, ChatGLM, and AutoClaw.

Product overview

Zhipu AI (also branded Z.ai; HKEX-listed as Knowledge Atlas Technology, stock code 02513) operates a platform-plus-modules architecture centered on its GLM family of foundation models. The core GLM large models — GLM-5.2 (flagship open-source 753B-parameter MoE), GLM-5V-Turbo (multimodal vision coding model), GLM-5-Turbo (agent-optimized), and AutoGLM (autonomous agent model) — form the GLM-OS ecosystem and are exposed commercially through the Bigmodel MaaS (Model-as-a-Service) API platform with enterprise-ready agent APIs (General Translation Agent, GLM Slide/Poster Agent), AI Web Search Tools, and Model Fine-Tuning services. These models power a portfolio of consumer and enterprise products: the Z.ai web chat platform, AutoClaw (first domestic local OpenClaw implementation), ChatGLM chatbot, Zlearn (learning), Zread.ai (reading comprehension), AMiner (academic mining), Autotyper, and the earlier CodeGeeX intelligent programming assistant and GLM-PC computer-use agent. The company is positioning itself as a leading independent Chinese LLM provider with a focus on coding capabilities, agentic AI, and open-weight distribution.

Differentiator

Problem solved

Functional benefit

Brands

• Z.ai: Consumer-facing web chat platform powered by GLM models
• AutoClaw
• ChatGLM
• Zlearn
• AutoGLM
• Zread.ai
• AMiner
• Autotyper
• Bigmodel

Products and services

• GLM-5.2 Flagship open-source 753-billion-parameter Mixture-of-Experts foundation model under MIT license with a 1M-token lossless context window, IndexShare sparse attention, and frontier coding performance (#2 on CodeArena, top of Design Arena, 81.0 on Terminal-Bench). Targeted at enterprise developers, engineering teams, and AI researchers needing frontier open-weight models.
• GLM-5V-Turbo Native vision-language coding model with multimodal fusion and joint reinforcement learning, optimized for visual development, UI understanding, and computer-use agent workflows on OpenClaw and Claude Code. For developers building UI-driven and computer-use agentic applications.
• GLM-5-Turbo GLM variant built for the OpenClaw agent ecosystem, optimized at the training level for tool calling, instruction following, and long-chain execution to power autonomous agent workflows. For developers building agentic AI applications on OpenClaw.
• AutoGLM Autonomous agent model capable of autonomous planning, reasoning, and execution with continuous self-improvement; can execute 50+ step long-horizon operations across apps and websites. For enterprises and developers needing automated cross-application task execution.
• Z.ai Web-based consumer chat platform offering free trials (20 million tokens), brainstorming, full-stack reasoning, magic design, information Q&A, search, content creation, and AI Slides features powered by GLM models. For individual developers and consumers.
• AutoClaw First domestic local OpenClaw implementation marketed as 'Every PC, 1 Minute,' enabling AI agents to operate autonomously across applications and devices on a user's PC. For PC users wanting on-device autonomous AI agents.
• ChatGLM Zhipu's conversational chatbot product (智谱清言), also deployed end-side on Intel Core Ultra (2nd-gen) AI PCs. For consumers and enterprise users wanting an AI assistant.
• Zlearn Learning-focused AI product leveraging the GLM family for educational use cases. For students and self-learners.
• Zread.ai AI-powered reading comprehension product for analyzing and understanding long-form text. For researchers, students, and knowledge workers.
• AMiner Academic mining and scientific search platform supporting knowledge retrieval for research workflows. For academic researchers and R&D teams.
• Autotyper Automated typing product for productivity use cases powered by GLM models. For end users wanting AI-assisted text input.
• Bigmodel Commercial Model-as-a-Service (MaaS) platform offering High-Performance Model APIs (GLM-5.2, GLM-5V-Turbo), Enterprise-Ready Agent APIs (General Translation Agent, GLM Slide/Poster Agent), and an End-to-End Development Suite (Model Fine-Tuning, AI Web Search Tools). For developers and enterprises integrating AI capabilities.
• CodeGeeX Intelligent programming assistant deployed end-side on Intel Core Ultra (2nd-gen) AI PCs as part of the Intel-Zhipu AI PC ecosystem partnership, supporting AI-assisted code generation. For developers on Intel AI PC hardware.
• GLM-PC Computer-use agent that operates computers like a human, capable of driving PC applications autonomously through screen-screenshot-based control. Base model CogAgent-9B open-sourced to drive the GLM-OS agent ecosystem. For enterprises needing automated computer-use workflows.

Quantifiable outcome

• ~131.9% YoY revenue growth in 2025 (724.3M yuan)
• +5 more outcomes

Named customers6 records

Segments5 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration10 records

AI capability12 records

Feature9 records

Partnerships

Twelve partnerships are on record, tiered flagship, core and minor.

• Huawei CloudflagshipTechnology or IntegrationMember of Huawei Cloud's AI Model Partner Program with more than 20 model providers. Zhipu's GLM models are distributed inside Huawei Cloud's Industry AI Foundry across Smart Healthcare, Embodied AI, Smart Manufacturing and Scientific Computing zones, with full-stack support for up to 100,000-accelerator-card Ascend clusters. Huawei Cloud CEO Peter Zhou cited Zhipu as a key partner for vertical productivity use cases in healthcare, embodied intelligence and confidential computing.
• IntelcoreOEM/ Whitelabel/ Licensing PartnerDeepening smart-ecosystem partnership announced at Intel's AI PC Summit (October 2024) covering on-device deployment of Zhipu's ChatGLM and the CodeGeeX AI coding assistant on Intel Core Ultra processors — a client-side AI PC distribution channel.
• China TelecomflagshipChannel Partner/ Reseller/ DistributorBundles Zhipu AI's GLM-5 alongside its proprietary TeleChat model inside nationwide token-priced AI packages — consumer tiers from 9.9–49.9 yuan/month and business tiers from 39.9–299.9 yuan/month — monetizing China's first telco-led "token economy."
• China MobilecoreChannel Partner/ Reseller/ DistributorCo-launches AI Token packages selling compute power for large AI models including Zhipu's as a new monetization model, alongside China Telecom and China Unicom.
• China UnicomcoreChannel Partner/ Reseller/ DistributorParticipates in the China telecom AI-token package rollout and is listed as a key investor in SiliconFlow alongside earlier backers Alibaba Cloud and Zhipu AI.
• Alibaba Cloud (Bailian MaaS)flagshipChannel Partner/ Reseller/ DistributorZhipu models are offered on Alibaba Cloud's Bailian Model-as-a-Service marketplace with switchable access and promotional lite (7.9 yuan) and pro (39.9 yuan) first-month pricing. Alibaba's low-cost AI coding-tools bundle also combines Qwen 3.5 with Zhipu AI, Moonshot AI and MiniMax models.
• Feishu (Lark)coreTechnology or IntegrationFeishu open-sourced the OpenClaw plugin and adopted Zhipu's models to operate cloud documents, multi-dimensional tables and schedules inside its enterprise workflow ecosystem.
• SiliconFlowcoreStrategic or Co-development PartnerSiliconFlow hosts and optimizes Zhipu's models alongside DeepSeek, Alibaba Qwen and Moonshot on a multi-model, multi-chip platform serving 10M+ users and 10,000+ enterprise customers; Zhipu AI is also an earlier backer of SiliconFlow's ~$300M funding round.
• OpenClaw ecosystem (Peter Steinberger / OpenAI)flagshipStrategic or Co-development PartnerZhipu released AutoClaw (first domestic local version) and GLM-5-Turbo as the recommended backbone for the viral OpenClaw AI-agent framework; multiple Chinese cities (Shenzhen, Wuxi, Hangzhou, Nanjing) have offered subsidies up to CNY 2M for OpenClaw-related development.
• Cyberspace Administration of China (CAC)coreOthersZhipu AI is among the Chinese AI developers (with DeepSeek, Alibaba, Tencent) required to file comprehensive safety assessments and registration documentation before releasing generative-AI products under China's 2023+ registry system (868 filings, 530 registered services as of April).
• Zhongguancun Independent Large Model Industry AlliancecoreOthersFounding alliance chaired by Zhipu Chairman Liu Debing, bringing together 32 member companies to coordinate domestic LLM industry development from Beijing's Haidian District.
• LEAP East (Tahaluf / Saudi MoCIT)minorGTM or Marketing PartnerZhipu AI is a confirmed corporate speaker at LEAP East Hong Kong (8–10 July 2026), which is backed by Saudi Arabia's Ministry of Communications and Information Technology and formally endorsed by the Hong Kong SAR Government's Innovation, Technology and Industry Bureau.

Scale indicators17 records

Recent moves10 records

Expansion highlights7 records

Direct peers

• DeepSeek: Hangzhou-based foundation model lab whose DeepSeek-V3.2 is the most directly comparable Chinese open-weight LLM competitor to Zhipu's GLM family; both target coding/agentic workloads, sell via MaaS APIs, and compete head-to-head in China Telecom bundles.
• Moonshot AI: Beijing-based Chinese LLM developer behind the Kimi chatbot and open-weight Kimi K-series models; competes with Zhipu on long-context reasoning, MaaS API pricing, and Alibaba Cloud Bailian marketplace placement.
• Alibaba (Qwen): Alibaba's Qwen 3.5 foundation model is bundled alongside Zhipu, Moonshot and MiniMax in Alibaba Cloud's Bailian MaaS marketplace and coding-tools bundle, making it a direct Chinese open-weight LLM rival with overlapping enterprise and developer customer base.
• StepFun (Step AI): Shanghai-based Chinese multimodal foundation model lab and co-member of Huawei Cloud's AI Model Partner Program; competes with Zhipu on step-reasoning, multimodal capabilities and Chinese enterprise/government deployment channels.
• SenseTime: Hong Kong-listed Chinese AI company developing multimodal foundation models and the SenseChat/SenseNova family; competes with Zhipu on enterprise multimodal deployments and is also a co-investor with Zhipu in Dexmal and SiliconFlow.
• OpenAI: US frontier foundation model lab behind GPT-5.5; the global benchmark for Zhipu's coding and long-horizon agentic comparisons and the anchor reference point for Zhipu's ~1/6 cost-advantage narrative.
• Anthropic: US frontier AI lab behind Claude Opus/Fable 5; directly comparable to Zhipu on long-horizon coding benchmarks, agentic engineering (Claude Code integration with GLM-5V-Turbo), and enterprise API pricing.

Broad incumbents

• ByteDance (Doubao/Seed): China's largest private tech company operates the Doubao LLM family and Volcano Engine MaaS platform, representing a broad Chinese incumbent in foundation models that competes with Zhipu across consumer, enterprise and government channels.
• Tencent (Hunyuan): Tencent's Hunyuan foundation model and Weixin-/QQ-distributed AI products are a broad Chinese incumbent comparable to Zhipu; Tencent was also an early Zhipu backer, making it both competitor and shareholder.

Emerging players

• Mistral AI: Paris-based open-weight European foundation model developer with comparable open-source distribution strategy and MaaS API business; a relevant emerging open-weights peer, though smaller in scale and Chinese-market reach than Zhipu.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat6 records

Key risks7 records

Key highlights7 records

Customer concentration

• Xx.com/zai_org
• GitHubgithub.com/THUDM
• LinkedInlinkedin.com/company/zdotai

Revenue estimate

Valuation estimate

Number of profiles

Profiles4 records

Funding overview

Funding rounds23 records

Investors31 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A2 records

Investments20 records

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Zhipu AI