Organisation · lab · CN

Zhipu AI

The open-weight GLM family of models, spun out of Tsinghua University; now listed as Z.ai.

19 entries · March 2023 – September 2026

Zhipu AI is a Chinese lab that grew out of Tsinghua University's Knowledge Engineering Group and builds the GLM family of models. It was an early mover in open Chinese models, releasing the bilingual ChatGLM-6B in 2023 — small enough to run on a single consumer graphics card — well before the larger open releases from Alibaba and DeepSeek that followed. Now trading publicly as Z.ai after a Hong Kong listing in early 2026, it releases frontier-scale models under permissive licences. Its 2026 GLM-5, reported to have been trained entirely on Huawei Ascend chips rather than Nvidia hardware, drew particular attention as a marker of China's effort to build an AI compute stack independent of US export controls; the company has been on the US Commerce Department's Entity List since January 2025.

Category
Chinese AI labs
Founded
2019
HQ
Beijing, CN
Key people
Zhang Peng

Appears alongside

Featured in threads

Tracks

  • Models & capabilities 9
  • Open weights & ecosystem 9
  • Security & misuse 5
  • Compute & infrastructure 4
  • Benchmarks & progress 4
  • Money & business 3
  • Safety & alignment 2
  • Government & policy 2

  1. Z.ai says GLM-5.3 built much of its own inference stack

    The company said an agent built on its own model tripled inference throughput on a 100,000-chip domestic cluster within two weeks, while insisting humans kept control of objectives and risk.

    Models & capabilities · Compute & infrastructure

  2. Study detects reward hacking from models' internal representations

    Cheap difference-of-means probes on frontier open-weight models matched expensive LLM-judge monitors at catching reward hacking, at a fraction of the compute cost.

    Moonshot AI, Alibaba / QwenSafety & alignment · Benchmarks & progress

  3. Zhipu raises $5bn in two months

    A Hong Kong share placement and zero-coupon convertible bonds followed a $4bn raise in July, taking money raised since January's IPO to roughly $9.6bn.

    Money & business · Compute & infrastructure

  4. Anthropic details how states and criminals misused Claude

    Anthropic's September threat report said Claude had been used to write missile-guidance software for a Yemen weapons cell and to support biological-weapons research, alongside autonomous cyberattacks, state surveillance and the Chinese distillation campaigns.

    Anthropic, Alibaba / QwenSecurity & misuse · Safety & alignment

  5. US agencies accuse six Chinese labs of distilling US AI models

    A joint NSA/CISA/FBI advisory named DeepSeek, Moonshot, Alibaba, MiniMax, StepFun and Z.AI as running systematic distillation campaigns against Claude, GPT, Gemini and Grok since late 2024; China's Commerce Ministry rejected it and threatened countermeasures.

    US Government, Chinese GovernmentSecurity & misuse · Government & policy

  6. Zhipu reports first-half 2026 revenue up sharply but below estimates

    Cloud and API sales grew to about 86.5% of revenue, up from roughly 15% a year earlier, as the company's net loss narrowed by about 12%.

    Money & business

  7. Zhipu reveals GLM-5.3-Flash after testing it anonymously online for a week

    Priced at roughly a tenth of GLM-5.2 and released under an MIT licence, it is the first natively multimodal model in the GLM-5 series.

    OpenRouterModels & capabilities · Open weights & ecosystem

  8. Zhipu releases GLM-5.3 with an unplanned jump in cyber capability

    Built on the same unretrained base model as GLM-5.2, it scored 84.5% on the CyberGym vulnerability-discovery benchmark and had its weights withheld for roughly two weeks.

    Models & capabilities · Security & misuse

  9. Mistral opens its platform to third-party models and pitches European AI infrastructure

    It began hosting a rival lab's model, Zhipu's GLM-5.2, alongside a pledge — backed by Amadeus, ASML and Capgemini among others — to build up to 1GW of capacity by 2030.

    Mistral AICompute & infrastructure · Open weights & ecosystem · Money & business

  10. UK AISI and US CAISI jointly assess Moonshot AI's Kimi K3 for cyber capability

    Kimi K3 failed to produce a working exploit on any of 41 code-execution tasks, versus 20 of 41 for the leading closed US models tested with safeguards disabled.

    UK AI Security Institute, US Center for AI Standards and InnovationSecurity & misuse

  11. UK AISI reports narrowing cyber-capability gap between open-weight and closed frontier models

    On a 70-task cyber suite, GLM-5.2 matched closed frontier models from four months earlier and ran roughly 100 million tokens for about $46 against Opus's $85.

    UK AI Security Institute, DeepSeekSecurity & misuse · Open weights & ecosystem

  12. US CAISI publishes assessment of Z.ai's GLM-5.2

    The US assessment found GLM-5.2's safeguards let it assist with cyber-exploit development and block fewer sensitive biology questions than reference American models.

    US Center for AI Standards and Innovation, OpenAIBenchmarks & progress · Open weights & ecosystem

  13. WSJ reports China has 'matched' Anthropic in cybersecurity; Zvi and others dispute the framing

    Critics said the report conflated finding vulnerabilities when pointed at them, which GLM-5.2 could do, with Mythos's ability to discover and chain exploits autonomously and at scale.

    AnthropicBenchmarks & progress · Government & policy

  14. GLM-5.2 becomes the leading open-weight model

    It ranked #25 overall on LMArena, #8 on EQ-Bench and second on Vending-Bench 2, but commentators noted it cost more per task than smarter closed rivals.

    Open weights & ecosystem · Models & capabilities

  15. Zhipu AI releases GLM-5.2, tops open-weight rankings

    The MIT-licensed, 744-billion-parameter model scored 51 on Artificial Analysis's Intelligence Index, the highest of any open-weight model, days after Washington forced Anthropic offline for foreign users.

    Open weights & ecosystem · Models & capabilities

  16. Zhipu (Z.ai) releases GLM-5, trained entirely on Huawei Ascend chips

    Released under the MIT licence, the 744-billion-parameter model scored 77.8% on SWE-bench Verified, days ahead of new Alibaba and ByteDance model launches.

    Alibaba / Qwen, MiniMaxModels & capabilities · Open weights & ecosystem · Compute & infrastructure

  17. Zhipu (Z.ai) releases GLM-4.5 series

    355B-parameter open-weight model family aimed at agentic use, part of China's open-source push after DeepSeek-R1.

    Open weights & ecosystem · Models & capabilities · Benchmarks & progress

  18. Zhipu launches GLM-4, claiming near-GPT-4 parity

    Unveiled at Zhipu's first DevDay with a 128K-token context window, alongside a $100 million fund the company pledged to LLM startups.

    Models & capabilities

  19. Zhipu open-sources ChatGLM-6B

    The 6.2-billion-parameter model, trained on roughly a trillion tokens of Chinese and English text, could run on a single consumer graphics card with 6GB of memory using INT4 quantisation.

    Tsinghua UniversityOpen weights & ecosystem · Models & capabilities

In the commentary

Pieces from around the web that discuss Zhipu AI. External links.

Also mentioned in 9 entries

Referenced in passing — Zhipu AI isn't the main subject of these.