12 questions answered. Every answer sourced.
GLM-5: Every Z.ai Model From GLM-5 to GLM-5.3, Explained
GLM is the model family from Z.ai, the Chinese lab formerly known as Zhipu AI. In 2026 it shipped five GLM-5 releases, including the open-weights model that NIST's evaluators called probably the most capable when it launched. Here is what each version is, what it costs and how to use it.
GPTPrompts.AI Editorial
Facts checked against official announcements and documentation, October 2026 · Last updated October 2026
Affiliate link, we may earn a commission at no extra cost to you.
01 · The lineup
Every GLM-5 model
Five releases in seven months, on two base architectures.
What is GLM-5?
GLM-5, released on February 12, 2026, is a mixture-of-experts language model with 744 billion parameters, 40 billion of them active per token, and a 200,000 token context window. It was the largest GLM model at launch and is MIT licensed. Every later GLM-5 point release is built on the same architecture. On Z.ai's API it is still sold at $1.00 input and $3.20 output per million tokens.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
What is the difference between GLM-5.2 and GLM-5.3?
GLM-5.2 (June 16, 2026) extended context to 1 million tokens, is trained for long-horizon tasks and is MIT licensed. GLM-5.3 (August 18, 2026) uses the same base model and price but improves coding, security and agent tasks; Z.ai's model card shows it beating GLM-5.2 on every shared test. The catch is the license: GLM-5.3 is the first GLM flagship since GLM-4.5 not released under MIT. Both are text only.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Z.ai: GLM-5.2, built for long-horizon tasks; Fello AI: GLM 5.3 benchmarks, pricing and the held-back weights
What is GLM-5.1?
GLM-5.1, released on April 7, 2026, kept GLM-5's 200,000 token context, is MIT licensed and introduced the $1.40 input and $4.40 output price that every later flagship has kept. If you call GLM-5.1 or GLM-5.2 through the GLM Coding Plan, Z.ai now routes the request to GLM-5.3.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
What is Ox Alpha?
Ox Alpha was the anonymous test name for GLM-5.3-Flash before launch. Z.ai told investors it processed 62 trillion tokens in its first six days. It was released officially on August 26, 2026 as a 320 billion parameter model with 18 billion active, a 1 million token context window, image input and a plain MIT license, and it is the default model in the chat.z.ai app.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Z.ai: GLM-5.3-Flash, frontier intelligence at flash cost
02 · Performance
How good is GLM?
The strongest open-weights family by several independent measures.
How good is GLM-5.3 compared with GPT and Claude?
Strong but not first. On the Artificial Analysis Intelligence Index, GLM-5.3 scored 44.9 in September 2026, the highest of any open-weights model and ahead of every other Chinese model, but about 8.5 points behind Claude Fable 5.1 and below GPT-6 Astra. Z.ai's own model card reports 88.2 on Terminal Bench 2.1; treat vendor scores as a starting point, not proof.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Fello AI: GLM 5.3 benchmarks, pricing and the held-back weights
Was GLM-5.2 really the best open model?
The US government's evaluators thought so. NIST's Center for AI Standards and Innovation (CAISI) described GLM-5.2 as probably the most capable open-weight AI model when it was released, with overall capability similar to GPT-5.2. The same assessment found its safeguards allowed help with agentic cyber exploit development, so test it before exposing it to untrusted input.
Source: NIST CAISI: Assessment of Z.ai's GLM-5.2; Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
03 · Pricing and access
How to use GLM and what it costs
Free chat, cheap API, and a flat plan for coding tools.
How much does GLM-5.3 cost?
On Z.ai's API, GLM-5.3 costs $1.40 per million input tokens, $0.26 per million cached input tokens and $4.40 per million output tokens. GLM-5.3-Flash costs $0.15 and $0.50, about a ninth of the flagship, and GLM-4.7-Flash is free. Third-party hosts such as OpenRouter list GLM from several providers at different prices, so compare before committing.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Fello AI: GLM 5.3 benchmarks, pricing and the held-back weights; Z.ai: API platform
What is the GLM Coding Plan?
A flat subscription that plugs GLM-5.3 and GLM-5.3-Flash into coding tools you already use, including Claude Code, Cursor, Cline, Codex, OpenCode, Roo Code, Kilo Code and TRAE, plus Z.ai's own ZCode. Lite costs $18 a month, Pro $80 and Max $168, with 20% off quarterly and 30% off yearly. Usage is metered in weekly credits, and requests outside the weekday peak of 14:00 to 18:00 Singapore time use half the credits.
Source: Z.ai: GLM Coding Plan; Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Fello AI: GLM 5.3 benchmarks, pricing and the held-back weights
Is GLM free to use?
Yes, in two ways. The chat.z.ai web app is free after sign-up, and some smaller models, including GLM-4.7-Flash, cost nothing on the API. The newest flagship models are paid.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
Can I run GLM locally?
Yes for the open-weights models. GLM-5.3-Flash, GLM-5.2, GLM-5.1 and GLM-5 publish weights on Hugging Face under MIT, and GLM-5.3's weights are public under a custom license. The full GLM-5 models need data-centre hardware; GLM-5.3-Flash, with 18 billion active parameters, and quantized community builds are the realistic options for a powerful workstation through tools such as Ollama or LM Studio.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026); Z.ai: GLM-5.3-Flash, frontier intelligence at flash cost
04 · The company
Who makes GLM, and what to check first
Licensing and compliance matter more than usual here.
Are Zhipu AI and Z.ai the same company?
Yes. Zhipu AI was founded in Beijing in 2019. Z.ai is its international brand, and its legal English name became Z.AI Co., Ltd. in July 2026. It listed in Hong Kong on January 8, 2026 under stock code 2513 and counts more than 7.4 million enterprise and developer users.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
Is it safe for my company to use GLM?
Check two things first. Zhipu and several affiliates have been on the US Commerce Department's Entity List since January 16, 2025, which restricts exports of US items to them, and Texas bans the company's technology on state agency devices. Self-hosting the MIT-licensed weights avoids sending data to Z.ai, but GLM-5.3's custom license requires a Z.ai security review for AI service companies above $10 billion in revenue. Ask counsel before building a product on it.
Source: Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
Sources
Read and checked October 2026. Policies change, so confirm anything time-sensitive on the source page.
- Second Talent: Every GLM (Z.ai) model explained and compared (Sep 15, 2026)
- Z.ai: GLM-5.2, built for long-horizon tasks
- Fello AI: GLM 5.3 benchmarks, pricing and the held-back weights
- Z.ai: GLM-5.3-Flash, frontier intelligence at flash cost
- NIST CAISI: Assessment of Z.ai's GLM-5.2
- Z.ai: API platform
- Z.ai: GLM Coding Plan