Z.ai profile, products and governance

Z.ai suits developers who want strong open-weight coding models at low API prices, with some releases under MIT and the newest flagship under its own permissive license. Skip it if you need audited compliance certifications, or must avoid suppliers on the U.S. Entity List.

Z.ai trades abroad as the name for Zhipu AI, a Beijing lab whose shares have been listed in Hong Kong since January 2026. Its privacy policy names a Singapore company as data controller, and its API terms say customer content is not stored. It sells GLM models through an API, a chat product and monthly Coding Plans.

Founded: 2019 · HQ: Beijing, China · Team: 2000+ · CEO: Zhang Peng · Funding: $1.996B pre-IPO, plus a $558M HK IPO (Jan 2026), a follow-on share sale of about HK$31.4B (Jul 2026), and a September 2026 package of about HK$15.7B in new shares plus a RMB 20.14B convertible bond · Valuation: $6.55B at IPO (Jan 2026); shares have traded from the HK$116.20 IPO price to a July 2026 high near HK$2,980 (HKEX: 2513)

About Z.ai

Is Z.ai safe to use? It depends on the product and on who is buying. Abroad, Zhipu AI, a Beijing lab, goes by Z.ai; it has been listed in Hong Kong as Knowledge Atlas Technology (HKEX: 2513) since 8 January 2026, so its owners and filings are public. Its privacy policy names a Singapore company, Jingsheng Hengxing Technology Pte. Ltd., as data controller and says the services are generally provided from Singapore. The data processing addendum for API customers says the company does not store the content that customers and their end users send or generate, and the Coding Plan page says Team plan data is never used for model training. What Z.ai has not disclosed is any SOC 2 or ISO 27001 certification. The U.S. Commerce Department also added Zhipu AI to its Entity List in January 2025, and we found no report of a removal as of 2026-10-13. The consumer privacy policy we read was last updated on 29 September 2025, so read the current text before sending sensitive data to the chat product.

Z.ai was founded in 2019 by Tang Jie and Li Juanzi, researchers from Tsinghua University's Knowledge Engineering Group, and rebranded internationally from Zhipu AI in July 2025. Its January 2026 IPO raised about $558 million at a $6.55 billion valuation, making it China's first listed frontier LLM developer; HokAI tracks it alongside dozens of other labs in its AI company directory. Two raises followed. The July 2026 follow-on sold 19.78 million shares at HK$1,588, about HK$31.4 billion, according to Silicon UK's reading of the filing. On 13 September 2026 the company filed for roughly 21.97 million new H shares at HK$714 (about HK$15.7 billion) plus a RMB 20.14 billion (about $3 billion) zero-coupon convertible bond. Counting the roughly $1.996 billion raised before the IPO, that is about $11.5 billion since 2023. The share price has been volatile: a July high near HK$2,980, then HK$721 on the Monday after the September filing, against an IPO price of HK$116.20. Local media have also reported a planned second listing on Shanghai's STAR Market, though no application had been publicly accepted when Silicon UK reported it.

The product that matters most is the GLM family, which traces back to GLM-130B in 2022. GLM-5.3 appears in Z.ai's release notes on 18 August 2026, and its weights were posted on Hugging Face on 25 August. It uses the same base model as GLM-5.2, with every gain coming from post-training, and the model page lists a 1-million-token context window, 128K maximum output and text-only input. Z.ai reports 28.3 on Terminal Bench 3.0 for GLM-5.3 against 33.7 for Fable 5 on its own model card, so treat it as strong open-weight coding rather than the closed frontier. Licensing is the detail most summaries get wrong: GLM-5.2 and GLM-5.3-Flash carry the MIT license on Hugging Face, but GLM-5.3 ships under its own GLM-5.3 License. It permits commercial use, modification and redistribution, with one condition: a company that runs a model-as-a-service business and earns more than $10 billion in any 12 months must pass a Z.ai security review first. GLM-5.3-Flash arrived on 26 August 2026 as a roughly 320-billion-parameter hybrid model with 18 billion active, priced at $0.15 input and $0.50 output per million tokens against $1.40 and $4.40 for GLM-5.3.

Z.ai sells access three ways: the chat product at z.ai, an OpenAI-compatible API documented at docs.z.ai, and the GLM Coding Plan subscription. As of 2026-10-29 the Coding Plan lists Lite at $18 a month (10,000 credits a week), Pro at $80 (6 times Lite) and Max at $168 (14 times Lite), with 20 percent off quarterly and 30 percent off yearly billing, plus a separate Team plan with seat management. The API pricing page also lists several free models, including GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash, and charges $0.01 per web search call. Because the weights are public on Hugging Face, inference hosts such as Fireworks AI and OpenRouter can serve the open-weight models without a Z.ai contract, and the Coding Plan advertises support for 20-plus agent tools including Claude Code and Cursor. For price context across vendors, see our cheapest LLM API guide and how to choose the right LLM.

Z.ai is led by CEO Zhang Peng, with co-founder Tang Jie as Chief Scientist and Liu Debing as Chairman. The company employs an estimated 2,000-plus people, mostly research staff in Beijing. It publishes model cards on Hugging Face and the GLM-5 technical report on arXiv (February 2026). During GLM-5.3's post-training it saw cyber capability rise faster than expected and published results on CyberGym and related benchmarks with the release; those figures are self-reported. Z.ai has not published a formal responsible scaling policy comparable to Anthropic's.

Where rivals win: a buyer who needs audited controls, or who must avoid suppliers on the U.S. Entity List, will likely do better with Google DeepMind, OpenAI and Anthropic, which publish compliance attestations Z.ai does not. Teams that need image or video input should note that GLM-5.3 is text-only. At home, Z.ai competes with DeepSeek, MiniMax, Moonshot AI and Alibaba's Qwen3.8-Max. Its low prices and permissive licensing are a deliberate bet on developer reach, but the pace of share and bond issuance since the IPO points to heavy training and compute spending rather than self-funding revenue. If you are still choosing between labs, Smart Match can shortlist options for your use case, and our guide to running AI locally covers when self-hosting open weights beats an API.

Mission

To build and openly distribute frontier AI models that serve developers and enterprises worldwide, starting with the open-weight GLM family.

Products

• GLM-5.3
• GLM-5.3-Flash
• Z.ai Developer API
• GLM Coding Plan
• AutoClaw

Links

• Website
• GitHub
• Twitter
• LinkedIn
• Blog
• Docs