Zhipu AI (Z.ai): The Company Behind GLM — Models, History, Stock and Products

Zhipu AI is the company now known as Z.ai, the developer of the GLM family of large language models. It was founded on March 15, 2019, originating from Tsinghua University research, and its model line runs from the original GLM (2021) and GLM-130B (2022) through ChatGLM to today’s GLM-5.3 and GLM-5.3-Flash. On January 8, 2026 it listed on the Main Board of HKEX under stock code 2513 as Knowledge Atlas Technology, and in July 2026 the listed company changed its name to Z.AI Co., Ltd.

If you have used GLM models, the chat.z.ai app, the Z.ai API or the GLM Coding Plan, you have used Zhipu AI’s products. This page covers what the company makes, how its names fit together, its history and stock listing, its open-source record and the latest Zhipu AI news, using only Z.ai’s own published information. To try the models right away, open the free GLM chat.

What is Z.ai (Zhipu AI)?

Z.ai is a foundation-model company. Its core product is the GLM family: large language models for chat, coding and agent work, plus vision, image generation, video, speech and OCR models. It sells access in three main ways: a free consumer chat app (chat.z.ai), a pay-as-you-go developer API (z.ai/model-api, documented at docs.z.ai), and the GLM Coding Plan, a subscription for AI coding tools. It also builds its own agent apps, ZCode and AutoClaw.

Two things define the company’s approach. First, it publishes open weights for most of its models, usually under the MIT license, so anyone can download and run GLM. Second, it ships fast: in 2026 alone it released GLM-5, GLM-5.1, GLM-5.2, GLM-5.3 and GLM-5.3-Flash, a new flagship roughly every two months. Its current flagship, GLM-5.3, is a 744B-parameter mixture-of-experts model (40B active) with a 1M-token context, and Z.ai describes it as the most capable open-weights model for coding.

Zhipu, Zhipu AI, ZhipuAI, Z.ai: the names explained

The company has used several names, and all of them still appear in search results, code and documentation. They all point to the same organization:

The practical takeaway: a Z.ai account and a bigmodel.cn account are separate, API keys from one do not work on the other, and “ChatGLM API” tutorials usually refer to the older platform. For new projects, use the Z.ai API at https://api.z.ai/api/paas/v4. The GLM API quickstart walks through getting a key, and Z.ai official website links lists every genuine domain.

Zhipu AI history: from GLM to GLM-5.3

Z.ai’s company page tells its early story as a series of model milestones. The later entries come from Z.ai’s API release notes.

Three phases stand out. From 2019 to 2023 the company built its research base: the GLM pre-training approach, the 130B-parameter GLM-130B and the ChatGLM line that made the name widely known. In 2024 it broadened into products around GLM-4: voice (GLM-4-Voice), agents (AutoGLM, and GLM-PC’s exploration of an “autonomous” PC) and a first reasoning model (GLM-Zero-Preview). From mid-2025 it moved to large mixture-of-experts models with open MIT weights and a fast release rhythm, starting with GLM-4.5 and reaching GLM-5.3 by August 2026. The full model-by-model record is on the GLM release timeline.

Zhipu AI stock: HKEX 2513

Zhipu AI is a publicly listed company. The key facts:

• Exchange: Main Board of HKEX.
• Stock code: 2513.
• Listing date: January 8, 2026.
• Name at listing: Knowledge Atlas Technology.
• Current listed name: Z.AI Co., Ltd. (changed in July 2026).
• Significance: at listing it was described as the first publicly listed foundation-model company.

If you are looking up the share price, search by the code 2513 on HKEX or your broker’s platform rather than by “Zhipu”, because the listed name changed in July 2026. This page does not track or quote share prices; they change every trading minute and nothing here is investment advice.

For anyone following the company as a business, the product signals are public and dated: the release cadence of flagship models, the pricing of the API (the GLM pricing page tracks it), the move of the GLM Coding Plan to credits-based tiers on July 30, 2026, and the open-weight releases on Hugging Face.

Z.ai products

GLM language models

The current line: GLM-5.3 (flagship, 1M context, $1.40 in / $4.40 out per 1M tokens), GLM-5.3-Flash (natively multimodal, $0.15 / $0.50), GLM-5.2, GLM-5.1, GLM-5 and GLM-5-Turbo, plus the GLM-4.x generation including the free GLM-4.7-Flash. The GLM models comparison puts all of them in one table.

Vision, image, video, speech and OCR models

GLM-4.6V and GLM-4.5V for visual understanding (with free GLM-4.6V-Flash), GLM-Image for image generation with strong text rendering ($0.015 per image), CogView-4 for images, CogVideoX-3 for video ($0.20 per video), GLM-ASR-2512 for speech recognition and GLM-OCR for document text extraction. There are also agents on the API: the GLM Slide/Poster Agent and a general-purpose translation agent.

chat.z.ai

Z.ai’s free consumer chat app at chat.z.ai. GLM-5.2 was available there on launch day, and GLM-5.3 is available there as well.

Z.ai API platform

The developer platform at z.ai/model-api with documentation at docs.z.ai. The API is OpenAI-compatible (Chat Completions at https://api.z.ai/api/paas/v4), with official SDKs for Python (zai-sdk) and Java. Features include thinking mode with reasoning effort control, function calling, JSON output, streaming, automatic context caching and a built-in web search tool.

GLM Coding Plan

A subscription for AI coding tools at z.ai/subscribe: Lite $18, Pro $80 and Max $168 per month, with discounts for quarterly and yearly billing and Team seats for organizations. It works inside tools such as Claude Code, Cursor, Cline, OpenCode, OpenClaw and Codex. Details on the GLM Coding Plan guide.

ZCode and AutoClaw

ZCode is Z.ai’s official agentic IDE and desktop app, described as an agent-powered coding environment optimized for GLM models. AutoClaw is an agent app for work tasks such as research, data analysis and automated workflows. Both are supported tools on the Coding Plan and are the focus of the GLM-5.3-Flash usage campaign running from September 3 to October 7, 2026.

AutoGLM and API agents

AutoGLM is Z.ai’s line of device-control agents, first released in October 2024. Its latest release, AutoGLM-Phone-Multilingual (December 11, 2025), is a multimodal mobile automation framework that reads screen content and performs real actions through ADB, and Z.ai says it handles natural-language tasks across more than 50 mainstream apps. On the API side, the GLM Slide/Poster Agent turns instructions into slides and posters ($0.70 per 1M tokens), and a general-purpose translation agent costs $3 per 1M tokens.

bigmodel.cn

The company’s older developer platform at bigmodel.cn and open.bigmodel.cn. It uses a separate account system from z.ai, so balances, keys and plans do not carry over between the two.

What Z.ai’s research is known for

Reading Z.ai’s launch posts and technical reports side by side, five technical themes keep coming back. They explain why the GLM line looks the way it does.

• Agentic engineering over chat. From GLM-4.5 (“native agentic LLM”) to GLM-5 (“from Vibe Coding to Agentic Engineering”) and GLM-5.1 (up to 8 hours on a single task), the stated goal is models that plan, use tools and finish long software tasks, not just answer questions.
• Efficient long context. GLM-5 integrated DeepSeek Sparse Attention; GLM-5.2 added IndexShare, which shares one indexer across every four sparse-attention layers and cuts per-token FLOPs by 2.9x at 1M tokens; GLM-5.3-Flash combined linear and sparse attention for a 4.4x smaller KV cache than GLM-5.3.
• Scaling reinforcement learning. Z.ai built slime, an asynchronous RL framework, to make post-training cheaper and faster. GLM-5.3 is the showcase: it uses the same base model as GLM-5.2, and every gain came from post-training.
• Security capability. Z.ai says cyber capability “developed faster than we expected” in GLM-5.3, which scored 84.5 on CyberGym in its published results. GLM-5.3’s weights were released two weeks after launch, once safety evaluation was complete, and its license adds a security-review clause for the largest model-hosting businesses.
• Cost-efficient multimodality. GLM-5.3-Flash starts from a new base model trained on a 30T-token multimodal corpus. Z.ai reports it approaches Claude Opus 4.8 overall at about one tenth of GLM-5.2’s price, and says the same recipe is being scaled to larger models.

The Z.ai developer ecosystem

Z.ai has built its platform to fit into tools developers already use, rather than asking them to switch:

• OpenAI-compatible API. Point the OpenAI Python SDK at https://api.z.ai/api/paas/v4/ and most existing code works with a new model ID.
• Anthropic-compatible endpoint. The Coding Plan exposes https://api.z.ai/api/anthropic so Claude Code and Goose can run on GLM models.
• Broad tool support. The Coding Plan works in ZCode, AutoClaw, Claude Code, Codex, Cursor, OpenCode, OpenClaw and more, with setup guides for Cline, Kilo Code, Roo Code, Goose, Crush and Factory in the docs.
• MCP servers. Every Coding Plan includes Vision Understanding, Web Search, Web Reader and Zread MCP servers.
• Third-party availability. GLM models are also listed on OpenRouter under the z-ai/ prefix, including a free, rate-limited GLM-5.2 variant.

Rate limits on the API are set per account and per model, and you can see yours at z.ai/manage-apikey/rate-limits. If you hit errors, GLM API rate limits and error codes explains each code.

Zhipu’s open-source record

Open releases have been part of the company’s strategy since GLM-130B in 2022. The record so far:

• GLM-130B (2022): open-sourced at more than 100B parameters.
• ChatGLM-6B (2023): downloaded more than 20 million times, according to Z.ai.
• GLM-4-9B and GLM-4V-9B (June 2024).
• GLM-4.5 and GLM-4.5-Air (July 2025) onward: flagship weights under MIT, continued through GLM-4.6, GLM-4.7, GLM-5, GLM-5.1 and GLM-5.2.
• GLM-4.7-Flash: the most-downloaded GLM repository on Hugging Face, with more than 1.8 million downloads a month.
• GLM-5.3-Flash (MIT) and GLM-5.3 (custom GLM-5.3 License with MIT-style permissions), both published on Hugging Face in August 2026.
• Tooling: slime, the asynchronous reinforcement-learning framework used to post-train the GLM-5 series, is open source, and the GLM-5 and GLM-4.5 technical reports are on arXiv.

Not everything is open: GLM-5-Turbo and the faster X and FlashX variants are API-only. The license for every model, and what the GLM-5.3 License adds, is explained in Is GLM open source?

Zhipu AI news: 2026 so far

The year’s major company and product events, newest first:

• September 2026: two Coding Plan promotions run at once. From September 3 to October 7, paid plan users get unlimited GLM-5.3-Flash in ZCode and AutoClaw between 23:00 and 09:00 UTC+8 and doubled quota on other agents. From September 25 to October 7, all plan usage is charged at the off-peak rate, half the standard credit rate.
• August 26: GLM-5.3-Flash, a new 320B-A18B base model and the first natively multimodal model of the GLM-5 series, released under MIT after testing anonymously as “ox-alpha”.
• August 14–25: GLM-5.3 announced (August 14), added to the API (August 18) and published with open weights (August 25). Z.ai reports a 50% gain over GLM-5.2 on its internal Z.ai Code Bench and state-of-the-art CyberGym results; working with security teams, the model found 2,436 vulnerabilities across 269 projects, which Z.ai tracks in its Security Disclosure Ledger.
• July 30: the GLM Coding Plan moves to credits-based tiers (Lite, Pro, Max). Legacy plans keep their terms until the end of the billing cycle.
• July: the listed company changes its name to Z.AI Co., Ltd.
• June 16: GLM-5.2 brings a 1M-token context and MIT weights on launch day.
• April 7: GLM-5.1, able to work on one task for up to 8 hours.
• March: GLM-5-Turbo for OpenClaw agent workflows, alongside the ZClawBench benchmark.
• February 12: GLM-5, 744B parameters, with a technical report titled “GLM-5: from Vibe Coding to Agentic Engineering”.
• January: HKEX listing (January 8), GLM-Image (January 14) and the free GLM-4.7-Flash (January 19).

For what may come next, Z.ai has made one official statement: in the GLM-5.3-Flash announcement it says it is “now scaling this recipe to larger models” and that the lessons “are already shaping our next frontier model”. There is no announced date or name for a GLM-5.5 or GLM-6.

How to use Zhipu AI’s models

1. Chat for free, no account: this site’s chat runs GLM-5.3-Flash by default and also offers GLM-5.3, GLM-5.2, GLM-5.1, GLM-4.7, GLM-4.6, GLM-4.5, GLM-4.7-Flash and GLM-Image. Try GLM-5.3 in the chat.
2. Z.ai’s own app: chat.z.ai, free.
3. The API: create a key at z.ai/manage-apikey/apikey-list and call https://api.z.ai/api/paas/v4/chat/completions. GLM-4.7-Flash is free per token.
4. In your coding tools: subscribe to the Coding Plan and point Claude Code or another supported tool at Z.ai’s endpoint (see GLM in Claude Code).
5. On your own hardware: download open weights from huggingface.co/zai-org and serve them with vLLM or SGLang.

A minimal API call looks like this:

curl https://api.z.ai/api/paas/v4/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $ZAI_API_KEY" \
  -d '{
    "model": "glm-5.3",
    "reasoning_effort": "low",
    "messages": [{"role": "user", "content": "What does Z.ai make?"}]
  }'

The same request with the OpenAI Python SDK, since the Z.ai API is OpenAI-compatible:

from openai import OpenAI

client = OpenAI(api_key="your-api-key", base_url="https://api.z.ai/api/paas/v4/")

resp = client.chat.completions.create(
    model="glm-5.3",
    reasoning_effort="low",
    messages=[{"role": "user", "content": "What does Z.ai make?"}],
)
print(resp.choices[0].message.content)

GLM-5.3 always thinks before answering, so keep reasoning_effort at low for quick questions and raise it to high or max for hard coding or analysis. Swap the model ID to glm-4.7-flash to run the same code at no token cost.

Official Z.ai links

Z.ai does not run a public status page, so if the service seems down, the steps in GLM not working? are the fastest way to find out whether the problem is on your side.

Zhipu AI FAQ

Is Zhipu AI the same as Z.ai?

Yes. Z.ai is the company formerly known as Zhipu AI. The GLM models, the ChatGLM line and the zai-org repositories all come from the same organization.

When was Zhipu AI founded?

March 15, 2019, according to Z.ai’s company page. It originated from Tsinghua University research.

Is Zhipu AI a public company?

Yes. It listed on the Main Board of HKEX on January 8, 2026 as Knowledge Atlas Technology, stock code 2513, and the listed company was renamed Z.AI Co., Ltd. in July 2026.

What is the Zhipu AI stock code?

2513 on HKEX. Search by the code rather than the name, since the listed name changed after the IPO.

What does Z.ai make?

The GLM model family (language, vision, image, video, speech and OCR models), the chat.z.ai app, a developer API, the GLM Coding Plan subscription, and the ZCode and AutoClaw agent apps.

What is Zhipu AI’s latest model?

GLM-5.3-Flash, released August 26, 2026, is the newest. GLM-5.3, announced August 14, 2026, is the flagship.

Are Zhipu AI’s models open source?

Most are open-weight under the MIT license. GLM-5.3 uses the custom GLM-5.3 License, and GLM-5-Turbo and the X/FlashX speed variants are API-only.

Does Zhipu AI have a free API?

Yes. GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash are free per token on the Z.ai API. Paid models range from $0.15 to $1.40 per 1M input tokens for the current generation.

Where is Zhipu AI on GitHub and Hugging Face?

Both use the organization name zai-org: github.com/zai-org and huggingface.co/zai-org. On ModelScope the organization is still called ZhipuAI.

What is the difference between bigmodel.cn and z.ai?

Both are Z.ai developer platforms, but bigmodel.cn is the older one and uses a separate account system. Keys, balances and plans from one do not work on the other.

The quickest way to judge what Zhipu AI has built is to use it: chat with GLM-5.3-Flash free, no sign-up.