GLM wrote it. Keep the decisions moving.

GLM, Z.ai and Zhipu

GLM is a series of large language models (LLMs) developed by Zhipu AI, based in Beijing, China. Zhipu runs the service in China, and internationally it offers chat (chat.z.ai), an API and a flat-rate coding plan under the "Z.ai" name. The international Z.ai service is operated by a Singapore company (JINGSHENG HENGXING TECHNOLOGY PTE. LTD.).

On Hugging Face, the models are published under the organization name "zai-org". News coverage mixes "Zhipu", "Z.ai" and "GLM", but they all refer to the same company's models and services.

Other things called "GLM"

The abbreviation "GLM" is used outside AI too. Adding a word to your search helps you find the right one.

• Generalized linear model: a statistical method that covers logistic and Poisson regression, common in R and Python statistics tutorials. Unrelated to the GLM on this page.
• GLM in the automotive world: sometimes used as a company name or model code. Also unrelated.
• GLM the AI model: Zhipu AI's model name, with a generation number and a variant, as in "GLM-5.3" or "GLM-4.7-Flash".

What GLM can do

• Chat: answering questions, writing, translating and summarizing.
• Coding: called from tools like Claude Code, Cline or OpenCode to write, fix and explain code.
• Image understanding and text extraction: GLM-5.3-Flash is multimodal and accepts images. There's also GLM-OCR for reading text from documents.

What sets it apart from ChatGPT's and Claude's flagship models is that near-flagship weights are published, so you can run them on your own servers or computer if you meet the conditions.

Scroll the table sideways →

Looking for "GLM-5.2"? As of October 2026, the flagship is GLM-5.3. You can still specify GLM-5.2 on the API, but on the GLM Coding Plan it's redirected to GLM-5.3. For older tutorials, read GLM-5.2 as GLM-5.3 and confirm against the official docs.

Model names change quickly, so it's enough to remember the roles: the flagship, the fast Flash, and the small one that runs locally.

The GLM Coding Plan lets you use GLM at a flat rate from coding tools such as Claude Code, Cline and OpenCode. Unlike the per-token API, you work within a monthly allowance, so costs are easier to predict if you code every day.

• Lite: $18 a month (as shown on the official page in October 2026; check tax treatment at checkout).
• Higher tiers: plans with larger allowances exist. See the official overview and the subscription page for details and prices.
• Allowance: each plan has usage limits (credits) per five hours and per week.
• Models: GLM-5.3 and GLM-5.3-Flash. Requests for GLM-5.2 go to GLM-5.3, and requests for GLM-4.7 go to GLM-5.3-Flash.
• Tools: Claude Code, Cline, OpenCode and other coding tools that let you change the endpoint.

Coding Plan or pay-as-you-go API?

• Coding Plan fits: individual developers who use coding tools every day and want a steady monthly bill.
• API fits: building it into your own systems, batch processing overnight, or usage that swings a lot month to month.

Neither handles "who reviews it" or "when it ships" after the code is written. If you're chasing that yourself every time, see how SIMY connects.

Z.ai offers an Anthropic-compatible API (https://api.z.ai/api/anthropic). Claude Code lets you change its endpoint with environment variables, so you can keep using Claude Code as usual with GLM running underneath.

1. Create an API key on Z.ai. If you're using the Coding Plan, subscribe first, then create the key.
2. Set the environment variables. Put the URL above in ANTHROPIC_BASE_URL and your API key in ANTHROPIC_AUTH_TOKEN. Never write the key directly into code or shared documents.
3. Start Claude Code. Per the official guide, Claude Code's Opus- and Sonnet-level calls map to GLM-5.3, and Haiku-level calls map to GLM-5.3-Flash. You can change the mapping with environment variables.
4. Try it on a small task. A different model means different strengths and a different response to instructions. Starting with something small, like fixing a test, is the safe way in.

Exact settings can change, so follow Z.ai's official instructions. Cline and OpenCode work the same way: set the endpoint and key.

How SIMY relates: SIMY doesn't connect GLM (Z.ai) API keys at this time. SIMY's cloud runs use your ChatGPT (Codex) account.

Scroll the table sideways →

Per million tokens, in US dollars, as listed on Z.ai's official pricing page in October 2026. Check the official page for tax treatment and the latest figures.

Getting started with the API

1. Create a Z.ai account. The international service is run by a Singapore company. Read the terms of service and privacy policy first.
2. Create an API key. Manage it in an environment variable.
3. Use the OpenAI-compatible or Anthropic-compatible endpoint. With existing SDKs and tools, swap the endpoint URL and model name.
4. Try it on the free Flash first. Check the flow with GLM-4.7-Flash, then switch to GLM-5.3 only where quality matters, to keep costs down.

License differences

You can't simply call GLM "open source". GLM-5.3-Flash uses the MIT license, which makes commercial use straightforward. GLM-5.3 uses its own "GLM-5.3 License", which is close to MIT in how much it allows, but API providers serving the model with very large revenue must pass a security review by Z.AI before commercial use. This usually doesn't affect an ordinary company using it internally, but read the license text before using it for work.

Running it on your own computer

• GLM-4.7-Flash: ollama run glm-4.7-flash in Ollama. The quantized version (q4_K_M) is about 19 GB and runs on a computer with plenty of memory.
• GLM-5.3 and GLM-5.3-Flash: Ollama's official library only has "cloud" tags. These run in Ollama's cloud, not on your machine.
• On your own servers: the GLM-5.3 weights are on Hugging Face, but at several hundred billion parameters you should plan for multiple large GPUs.

"Local" doesn't always mean "stays in-house": if the goal is keeping data inside the company, cloud-tagged models don't qualify. What runs on your own machine is a small model like GLM-4.7-Flash.

Work that suits running locally

Summarizing documents you can't send outside, working offline, and high-volume routine drafting every day are where local models shine. Small models aren't as capable as large ones, so use them for first drafts, not final versions, and you'll be less disappointed.

Ask GLM in your language and it will usually answer in kind. But the official model cards on Hugging Face list only English and Chinese, so quality in other languages isn't officially guaranteed.

• Good fit: generating and fixing code, summarizing English or Chinese material, drafting technical explanations.
• Worth checking: tone and formality, local names and terms, and anything about local laws or business customs. Always reread text that goes outside the company.
• With small models: small models run locally can produce awkward phrasing in long passages, especially outside English and Chinese.

The surest way to judge quality is to try a few real documents from your own work.

Scroll the table sideways →

Government action

On January 16, 2025, the US Department of Commerce's Bureau of Industry and Security (BIS) added Zhipu AI (Beijing Zhipu Huazhang Technology) and affiliated companies to the Entity List. Under this measure, exporting products or technology subject to US export regulations to these companies requires a license. Whether you can use it for work is a call to make under your organization's and your clients' rules.

• Check your company's rules first: some companies' generative AI policies limit which services you can use and what you can enter.
• Keep confidential data out by default: with any AI, think twice before entering customer information or unpublished figures.
• Run it locally if data must stay in: if you need to keep data inside the company, choose a model that runs on your own machine.

Data handling is based on Z.ai's privacy policy as of October 1, 2026. How SIMY handles data is published on our Security page and in our Privacy Policy.

Scroll the table sideways →

ByteDance's Doubao doesn't publish the weights of its flagship Seed models, and its API is split between Volcano Engine for China and BytePlus for international users. See the Doubao guide for details.

How they fit together: any AI, GLM included, can write the code and drafts. If you're chasing the owners, deadlines, messages and updates yourself every time after that, that's where SIMY comes in. See Chinese LLMs compared for an overview, or the Kimi, DeepSeek, Qwen and MiniMax guides.