Every GLM (Z.ai) Model Explained and Compared [2026]

TL;DR: Z.ai's flagship is GLM-5.3, released on August 18, 2026 at $1.40 per million input tokens and $4.40 per million output tokens. GLM-5.3-Flash is the cheap open model at $0.15 and $0.50 under an MIT licence, and GLM-4.7-Flash costs nothing on the API. Developers who want GLM inside Claude Code or Cursor can buy the GLM Coding Plan from $18 a month.

Z.ai earned RMB825 million from its open platform and GLM API in the first half of 2026, up 2,735.7% on a year earlier. Over the same six months its average API selling price rose about 101%, and API sales grew to 86.5% of company revenue.

The GLM Models at a Glance

Z.ai’s pricing page lists 19 GLM text and vision models today, from the free GLM-4.7-Flash to the GLM-5.3 flagship. Most have open weights on Hugging Face. The table covers the current text models and the vision model most teams use.

Model	Released	Size (total / active)	Context	Licence
GLM-5.3	Aug 18, 2026	GLM-5 architecture	1M	Custom GLM-5.3 licence
GLM-5.3-Flash	Aug 26, 2026	320B / 18B	1M	MIT
GLM-5.2	Jun 16, 2026	GLM-5 architecture	1M	MIT
GLM-5.1	Apr 7, 2026	GLM-5 architecture	200K	MIT
GLM-5	Feb 12, 2026	744B / 40B	200K	MIT
GLM-4.7	Dec 22, 2025	GLM-4.5 architecture	200K	MIT
GLM-4.7-Flash	Jan 19, 2026	30B / 3B	200K	MIT
GLM-4.6	Sep 30, 2025	GLM-4.5 architecture	200K	MIT
GLM-4.5	Jul 28, 2025	355B / 32B	128K	MIT
GLM-4.5-Air	Jul 28, 2025	106B / 12B	128K	MIT
GLM-4.6V	Dec 8, 2025	106B	128K	MIT

GLM-5.3: The Flagship

Z.ai shipped GLM-5.3 on the same base model as GLM-5.2 and at the same price: $1.40 input and $4.40 output. Z.ai’s model card compares the two on long coding and security tasks, where GLM-5.3 beats GLM-5.2 on every shared test, by 4.7 points on Agents’ Last Exam up to 30 points on ExploitBench.

On Artificial Analysis, GLM-5.3 trails Claude Fable 5.1 by 8.5 points and leads every other Chinese model. On the Arena text leaderboard, glm-5.3-max ranks 19th with 1,483 points, two places behind Moonshot’s Kimi K3.

Z.ai’s own model card reports 88.2 on Terminal Bench 2.1, against 88.8 for GPT-5.6 Sol, and 84.5 on CyberGym. Treat those as vendor figures: the card does not report SWE-bench Verified at all.

GLM-5.3 is text only. For images, screenshots or documents, use GLM-5.3-Flash or GLM-4.6V.

GLM-5.3-Flash: Open, Multimodal and Cheap

Under its test name “Ox Alpha”, GLM-5.3-Flash processed 62 trillion tokens in its first six days, Z.ai told investors in its interim results. It is the first GLM-5 model with image input, and it is the default model in the chat.z.ai app.

At $0.15 input and $0.50 output per million tokens, it costs about a ninth of GLM-5.3. It ranked fourth among all models on OpenRouter in the week to September 14, with 11.65 trillion tokens. Z.ai’s Flash launch post reports 84.3 on Terminal Bench 2.1 and 59.1 on OSWorld 2.0, a computer-use test.

The weights ship under plain MIT, so GLM-5.3-Flash is the model to self-host if you want the newest GLM without the custom licence. It activates 18B parameters per token, against 40B for GLM-5, so each token needs less compute.

GLM-5, GLM-5.1 and GLM-5.2

GLM-5 arrived on February 12, 2026 with 744 billion parameters, 40 billion of them active per token. It was the largest GLM model at release. Z.ai still sells it at $1.00 input and $3.20 output, the cheapest route to the GLM-5 architecture.

• GLM-5.1 (April 7, 2026) kept the 200K context and is listed at $1.40 and $4.40, the price every later flagship has kept.
• GLM-5.2 (June 16, 2026) extended context to 1 million tokens and stayed under MIT. NIST’s CAISI called it “probably the most capable open-weight AI model when it was released”, with overall capability similar to GPT-5.2.
• GLM-5-Turbo and GLM-5V-Turbo, two API-only variants priced at $1.20 and $4, disappeared from Z.ai’s pricing page between August 24 and August 29, 2026.

If you call GLM-5.2 or GLM-5.1 through the GLM Coding Plan, Z.ai now routes the request to GLM-5.3. On the pay-as-you-go API both models are still listed at the same price as GLM-5.3, so there is no cost reason to stay on them.

GLM-4.5, GLM-4.6 and GLM-4.7

GLM-4.5 set the pattern the lineup still follows: a large mixture-of-experts model under MIT, with a smaller Air sibling released the same day. It shipped on July 28, 2025 with 355B parameters and 32B active, next to the 106B GLM-4.5-Air.

Two point releases followed on the same architecture. GLM-4.6 (September 30, 2025) raised context to 200K tokens. GLM-4.7 (December 22, 2025) scored 73.8% on SWE-bench Verified by Z.ai’s count, the last time the company published that benchmark. Both cost $0.60 input and $2.20 output today.

The GLM-4 line also carries Z.ai’s free models. GLM-4.7-Flash, a 30B model with 3B active, is free on the API, as are GLM-4.5-Flash and the vision model GLM-4.6V-Flash. Two proprietary variants, GLM-4.5-X at $2.20 and $8.90 and GLM-4.5-AirX at $1.10 and $4.50, are API-only versions of the same pair. We reviewed the older GLM-4.6 in our GLM-4.6 coding review and against Claude in GLM vs Claude Sonnet.

Vision, Image, Speech and Phone Models

Z.ai’s API also covers vision, documents, images, video, speech and phone control, most of it cheap enough to run on every file or call:

Z.ai also open-sourced GLM-TTS, a text-to-speech model, under MIT in December 2025. It has no API price yet. The prices above are from Z.ai’s pricing page.

GLM Pricing Compared Across the Lineup

The most expensive GLM model on the API is not the flagship. GLM-4.5-X charges $8.90 per million output tokens, twice GLM-5.3, for an API-only version of a 2025 model.

Cached input is where GLM gets cheap for agents that resend long prompts. GLM-5.3 charges $0.26 per million cached tokens, and Z.ai lists cache storage as free for a limited time. GLM-5.3-Flash charges $0.03.

Against rivals, GLM-5.3 sits in the middle of the Chinese flagships. Kimi K3 costs $3 and $15, Qwen3.8-Max $2 and $6, and DeepSeek V4-Pro $1.32 and $3.96 at peak. Our US vs China AI model comparison sets these against GPT-6 and Claude, and the Kimi and DeepSeek guides cover those lineups.

Context Windows and Output Limits

ChatGLM-6B, the open model that started the line in March 2023, read 2,000 tokens. GLM-5.2 and everything after it reads 1 million.

Output limits moved less. GLM-4.6, GLM-4.7 and every GLM-5 model can write up to 128K tokens in one reply, and GLM-4.5 tops out at 96K. The vision model GLM-4.5V is the outlier, with a 64K context and 16K output.

The GLM Coding Plan

The GLM Coding Plan is a flat monthly subscription that plugs GLM-5.3 and GLM-5.3-Flash into coding tools you already use. Z.ai lists Claude Code, Cursor, Cline, Codex, OpenCode, Roo Code, Kilo Code and TRAE, plus its own ZCode.

Plans cost $18 (Lite), $80 (Pro) and $168 (Max) a month, with 20% off quarterly and 30% off yearly. Z.ai meters usage in credits: Lite gets 10,000 a week, Pro 60,000 and Max 140,000. Off-peak requests use half the credits. Z.ai’s docs define peak as 14:00 to 18:00 Singapore time on weekdays.

Prices have gone up this year. An April 21, 2026 notice listed Pro at $72 and Max at $160, and the interim report confirms the increase. Team seats cost $88 or $188 a month, and Z.ai says team data is not used for training by default.

Licences, the Entity List and Other Things to Check

Every GLM flagship from GLM-4.5 to GLM-5.2 shipped under MIT. GLM-5.3 did not, so check which side of the line your model sits on.

The US Commerce Department added Beijing Zhipu Huazhang Technology and several affiliates to the Entity List on January 16, 2025. The rule said they advance China’s military modernization through AI research. The listing controls exports of US items to those companies, so check it with counsel before you sell or license technology to Z.ai.

State rules can bite too. Texas added Zhipu (Z.ai) to its prohibited technology list for state agencies on January 26, 2026. On safety, CAISI found that GLM-5.2’s safeguards allowed help with agentic cyber exploit development. Run your own red-team prompts before a GLM model handles untrusted input.

Which GLM Model Should You Use?

• Long coding or agent runs: GLM-5.3. It is the strongest GLM model, and its price matches GLM-5.2.
• High-volume chat, extraction or image input: GLM-5.3-Flash, at about a ninth of the flagship’s price.
• Self-hosting with no licence questions: GLM-5.3-Flash or GLM-5.2, both MIT.
• Prototypes and hobby projects: GLM-4.7-Flash, free on the API.
• Scanned documents: GLM-OCR at $0.03 per million tokens, before a text model reads the result.
• Daily coding in Claude Code or Cursor: the GLM Coding Plan rather than pay-as-you-go, if you code most working days.

For how GLM compares with the other open Chinese models, see our ranking of Chinese open-source LLMs and the every Qwen model guide.

The Company Behind GLM

Z.ai began as Zhipu AI, founded in 2019, and changed its legal English name to Z.AI Co., Ltd. in July 2026. It trades in Hong Kong under stock code 2513.

Hong Kong retail investors subscribed the IPO’s public tranche 1,159 times over, according to Z.ai’s allotment announcement. The September placing price was a 9.96% discount to a HK$793 close, which is more than six times the IPO price.

The API business is now almost the whole company. In the six months to June 30, 2026, revenue rose 399.7% to RMB953.9 million, and the enterprise model business fell 54.6%, per the interim results. Z.ai still lost RMB2.07 billion over the half, and spent RMB2.13 billion on research. It counts over 7.4 million enterprise and developer users on its platform. Our list of top Chinese AI companies covers its rivals.

Build on GLM With the Right Engineers

Running GLM well means serving a mixture-of-experts model, wiring it into agent tools and checking its licence and safety before launch. Second Talent matches companies with vetted engineers from Asia who do this work. Hire LLM engineers, or find LLM engineers in China who already work with Chinese model stacks.

Tell us the role you need to fill and get a shortlist within 24 hours.

Frequently Asked Questions

Is GLM open source?

Most of it. GLM-5.3-Flash, GLM-5.2, GLM-4.7, GLM-4.6 and GLM-4.5 publish weights under MIT on Hugging Face. GLM-5.3’s weights are public too, but under a custom licence.

Is GLM free to use?

Yes, in two ways. The chat.z.ai app is free after sign-up, and GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash cost nothing on the API. The newer models cost money.

Are Zhipu AI and Z.ai the same company?

Yes. Zhipu AI is the original name of the Beijing company. Its legal English name became Z.AI Co., Ltd. in July 2026, and Z.ai is its international brand for GLM.