Skip to content
latestai.eulatestai.eu
G5

GLM 5.3

by Z.ai

Try itWebsiteGitHubHugging Face

Z.ai's open-weights 753B MoE language model, a post-trained GLM-5.2 aimed at coding, agentic tasks and cybersecurity.

Available in EULanguage ModelsModel

Updated EU status checked Last shown

GLM 5.3 at a glance

What is GLM 5.3?
GLM 5.3 is a model from Z.ai. Z.ai's open-weights 753B MoE language model, a post-trained GLM-5.2 aimed at coding, agentic tasks and cybersecurity.
Who makes GLM 5.3?
GLM 5.3 is made by Z.ai.
Does GLM 5.3 work in the EU?
Available in the EU. Weights are downloadable from Hugging Face by anyone and can be self-hosted in the EU; no EU exclusion was found in the provided material, though the custom glm-5.3 license terms were not reviewed. Hosted Z.ai service regional terms and EU data residency were not stated. Last verified 14 Nov 2026.
How much does GLM 5.3 cost?
Open source. Weights free to download under the glm-5.3 license. Hosted access via Z.ai chat and the Z Code plan; plan prices not given in the provided material. Creators describe it as very cheap. Licence: GLM-5.3 license (custom).
Where was GLM 5.3 shown?
In 8 creator videos by Matthew Berman, AI Search. Most recently by Matthew Berman on 11 Sept 2026, at 0:58 in “Deepseek did it again...”.

Sources: the vendor's own pages (linked above) and creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.

About

GLM-5.3 is a text-only large language model from Z.ai (Zhipu). It uses the same base model as GLM-5.2, and all gains come from additional post-training. The Hugging Face card lists a mixture-of-experts architecture (glm_moe_dsa), with English and Chinese as the stated languages. Creator coverage cites about 753B parameters and no native vision.

According to the model card, GLM-5.3 improves on GLM-5.2 mainly in complex coding and long-horizon tasks. Z.ai reports a 50% gain on its in-house Z.ai Code Bench and open-source state of the art on Terminal Bench 3.0 and Agents' Last Exam. It also reports a large jump in cyber capability: top CyberGym score (84.5) and more than double GLM-5.2 on exploitation benchmarks. The card's comparison table shows it trailing the best closed models (e.g. GPT-5.6 Sol, Fable 5) on several coding benchmarks while leading on AutomationBench and GDPval-AA. These figures are vendor-reported.

The weights are downloadable from Hugging Face under a custom "glm-5.3" license, including an FP8 variant. They can be served with SGLang, vLLM, Transformers, KTransformers, Unsloth and others. Community 1-bit GGUF quantizations from Unsloth are reported at around 217GB, so full deployment needs substantial hardware. It is also available through Z.ai's chat (which currently lists GLM-5.3-Flash) and the Z Code plan.

Key features
  • Open-weights ~753B mixture-of-experts text model
  • Post-trained from GLM-5.2 base with focus on coding and long-horizon agent tasks
  • Reported state of the art on CyberGym and open-source SOTA on Terminal Bench 3.0
  • Supports local serving via SGLang, vLLM, Transformers, KTransformers, Unsloth
  • English and Chinese; no native vision input
8 moments

As seen on YouTube

Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.

AI
AI Search17 Aug 2026
Live demo
New #1 open source AI has reached FRONTIER

New open-source frontier model from Z.ai, same architecture as GLM 5.2 but post-trained harder. The creator tested it on a Windows 11 web replica, a Blender V8 engine, a 3D fighting game, music composition, a financial video, deep research and vision tests. It has no native vision, scores top marks on GDPval, Automation Bench and Cyber Gym, and weights are promised two weeks after launch.

0:00Open on YouTube
Keep exploring

Similar AI tools

Closest neighbours by meaning, not just by category.

G5

GLM 5.3 Flash

Model · Zhipu AI

6

Zhipu AI's GLM-5.3-Flash, a fast open-weights LLM for coding and agentic tasks, available via Z.ai chat and API.

Available in EULanguage Models
Release date n/aFreemium
Q3

Qwen 3.8 Max

Model · Alibaba

8

Alibaba's open-weights 2.4T-parameter MoE language model (95B active) from the Qwen3.8 family, with a hosted Max version.

Available in EULanguage Models
Release date n/aOpen source
G5

GLM 5.2

Model · Zhipu AI

10

Zhipu AI's GLM 5.2 large language model family, with Fast and max variants, available via Z.ai chat and API.

UnverifiedLanguage Models
Release date n/a
IQ

IQuest Q1

Model · IQuest Lab

1

320B-parameter MoE model (15B active) from IQuest for agentic coding and tool use, with 512K context.

Available in EUCoding
28 Sept 2026Open source
O1

Ornith 1.5

Model

1

Open model family (9B dense, 35B MoE, 397B MoE) trained with a self-improvement loop for reasoning, coding and agentic tasks.

Available in EULanguage Models
Release date n/aOpen source
L2

LongCat 2.5

Model · Meituan

2

Meituan's open-weights 1.6T-parameter multimodal MoE model (48B active) for agentic coding and visual tasks.

Available in EULanguage Models
4 Nov 2026Open source