11 models · on the curve since 2023

Zhipu (Z.ai)

Value position · the site’s read Mixed

Zhipu’s durable position is the open GLM ecosystem plus the enterprise and state relationships of the first publicly listed Chinese AI lab. Public-market funding buys the runway; the open weights buy the adoption that compounds into it.

What would change this readOpen GLM loses developer mindshare to Qwen or DeepSeek, or the IPO capital fails to convert into a durable enterprise base.

Recent cadence
6 in last 12 mo
2 the year before
Landmark density
5 / 11
defining releases
Weights
Open weights
GLM-5.2 flagship

Founded 2019Beijing, China

The substrate underneath figures as of Aug 2026

Trains and serves on

Huawei Ascend, with an important limit on what is actually confirmed. GLM-Image (Jan 2026) completed its full training cycle on Huawei Ascend Atlas 800T A2 servers using MindSpore, which Zhipu presented as the first state-of-the-art multimodal model trained wholly on Chinese-made chips. The same claim for the GLM-5 language line is a Reuters report that Zhipu has not confirmed.

ConcentrationA US Entity List designation carrying Footnote 4 bars Zhipu from acquiring US-origin or US-technology-derived accelerators without a licence, leaving Huawei Ascend the only accelerator family it has publicly confirmed training on.

Flagship, per million tokens

$1.40 in $4.40 out

GLM-5.2, on the international platform

Was

$1.00 / $3.20

GLM-5 · Feb 2026

Weights
Open on the flagship line. GLM-5.2 ships under a plain MIT licence with no regional limit and no revenue threshold, unusually permissive for a frontier-class model; several Turbo and Flash API SKUs have no published weights.
Where you buy it
z.ai and bigmodel.cn · GLM Coding Plan · HuggingFace · Amazon Bedrock · Microsoft Foundry · OpenRouter
Structure
Beijing Zhipu Huazhang Technology, listed on the Hong Kong exchange as 2513.HK since 8 Jan 2026 — the first publicly listed company built around LLM foundation models. A 2019 spin-out from Tsinghua University. A second STAR Market listing is in progress. State-linked investors on the public record include Beijing, Hangzhou and Shanghai funds and the National Social Security Fund.
Raised
Listed January 2026 after an IPO raising about HK$4.3B; earlier private backers include Alibaba, Tencent, Meituan and Saudi-backed Prosperity7
Valuation
Market capitalisation passed HK$1 trillion, about US$128B, on 22 Jun 2026
Revenue
RMB 724.33M, about US$104.8M, for the year ended Dec 2025, up 131.9%
People
Under 900 at the January 2026 IPO, roughly three-quarters researchers

Biggest single dependencyIts training roadmap depends on Huawei Ascend supply and Chinese domestic fabrication capacity, because the Entity List designation closes the licensed route to US accelerators.

Two widely repeated claims are excluded here for lack of a source. The all-Huawei training claim is confirmed only for GLM-Image, not for the GLM-5 language line, where it is a Reuters report the company has not confirmed. The often-quoted 100,000-chip Ascend cluster traces to no credible source at all. GLM-5.2 parameter counts differ between aggregators and are absent from the model card, so none is given. Market capitalisation moves daily and quotes after 22 June are irreconcilable, so that dated figure is the only one shown. Prices here have RISEN twice in 2026, against the direction of the curve.

On the durability curve frontier era
2023202420252026 ChatGLM-6BGLM-5.2 ARCHITECTURE the substrate · compounds  → OUTPUT per model · commoditises  → CROSSOVER the durability thesis  → DURABLE VALUE-CAPTURE ↑ the margin a release keeps · not raw capability TIME · FRONTIER MODEL RELEASES →
Output · value a model keeps Architecture · the compounding substrate Zhipu (Z.ai) release

Hover any point to read the release. The y-axis is qualitative; only the x-axis carries dates.

When to reach for them practical read
Open agentic flagship
Reach for GLM-4.5 for open agentic quality (Jul 2025).
Vision-language
Pick GLM-4.5V when the task is multimodal (Aug 2025).
MIT-licensed
Use GLM where a permissive MIT license matters (GLM-5, post-IPO, Feb 2026).
The landmark releases 5 of 11 models
  1. Jun 2026 GLM-5.2 Latest 753B-parameter MoE under an MIT licence, 1M context OpenMIT1M ctx
  2. Feb 2026 GLM-5 744B MoE, first post-IPO flagship Open744B MoEPost-IPO
  3. Jul 2025 GLM-4.5 Open agentic flagship OpenAgentic
  4. Jan 2024 GLM-4 Flagship generation Flagship
  5. Mar 2023 ChatGLM-6B First open GLM chat model OpenFirst GLM
Full release logall 11, newest first
  1. 2026
  2. Jun GLM-5.2 753B-parameter MoE under an MIT licence, 1M context OpenMIT1M ctx
  3. Apr GLM-5.1 Open-source point release Open
  4. Feb GLM-5 744B MoE, first post-IPO flagship Open744B MoEPost-IPO
  5. 2025
  6. Dec GLM-4.7 Year-end flagship OpenFlagship
  7. Sep GLM-4.6 Coding and agent upgrade OpenCoding
  8. Aug GLM-4.5V Vision-language model OpenVision
  9. Jul GLM-4.5 Open agentic flagship OpenAgentic
  10. 2024
  11. Aug GLM-4-Plus Upgraded flagship Flagship
  12. Jan GLM-4 Flagship generation Flagship
  13. 2023
  14. Oct ChatGLM3 Third-gen open release Open
  15. Mar ChatGLM-6B First open GLM chat model OpenFirst GLM
Where they sit on weights versus the field
The durability take on Zhipu (Z.ai) from the archive
About Zhipu (Z.ai) at a glance
Founded
2019
HQ
Beijing, China
Leadership
Zhang Peng (CEO)
Origin
Tsinghua spin-off
Flagship
GLM-5.2
Weights
Open (MIT)
Listed
HK IPO, Jan 2026
Also builds
GLM, ChatGLM

Zhipu AI, operating internationally as Z.ai, was founded in 2019 as a Tsinghua University spin-off and is headquartered in Beijing. It builds the GLM family and became the first publicly listed Chinese AI lab via its Hong Kong IPO in January 2026, releasing flagship GLM models as open-weight under MIT.

The frontier chart → Read the archive →