GLM-5.2

GLM-5.2

A 753B-parameter open text model from zai-org for long, multi-step work such as coding and agent tasks, with a 1M-token context and an MIT license. For teams with serious GPU clusters.

Made by zai-org on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: GLM-5.2 is the latest flagship text model from zai-org, built for long-horizon tasks. The card says it holds up over a 1M-token context and improves on GLM-5.1 in coding. The record lists 753B parameters, English and Chinese, and an MIT license.

Who it suits: Teams with large GPU clusters who want an open-weight model for coding, reasoning and agent work, and developers who use it through the hosted API.

Industry | Who uses it | How they use it | Example
Software | Developers | Use it for coding and agent tasks | A repository-wide change from one request
Enterprise | Platform teams | Host an open model with an MIT license | An internal assistant on private servers
Research | AI labs | Study a large sparse-attention model | Compare it with GLM-5.1
Legal and finance | Analysts | Work through long documents | A long filing read in one context

You can skip it if you have no multi-GPU server. The weights are about 1,507 GB in BF16, so the hosted API is the lighter way in.

Where to use it:

• SGLang, vLLM, Transformers and KTransformers, with the versions the card lists
• Unsloth, from a beta version the card names
• Ascend NPU platforms through vLLM-Ascend, xLLM and SGLang
• The Z.ai API Platform, as a hosted service

Use case | Who | What they get back
Coding | Developers | Code written and edited at a chosen effort level
Long context work | Analysts | Answers across a context of up to 1M tokens
Agent tasks | Developers | Multi-step tool and terminal work

How to set it up:

• Python and pip on your computer
• A multi-GPU server, since the weights are about 1,507 GB in BF16 and about 377 GB at 4-bit by estimate
• One of the frameworks the card lists, at a version it names

1. Install the Hugging Face hub package.
2. Download the model files with the hf command.
3. Pick a framework from the card and follow its cookbook or recipe.

pip install -U huggingface_hub
hf download zai-org/GLM-5.2

How to use it: Serve the model with one of the listed frameworks, then send chat requests. The card says GLM-5.2 offers several thinking effort levels so you can trade quality against speed.

Framework | Minimum version on the card
SGLang | 0.5.13.post1
vLLM | 0.23.0
Transformers | 0.5.12
KTransformers | 0.5.12
Unsloth | 0.1.47-beta

Example: You send it a coding task with a long codebase pasted in and pick a high effort level. It works through the steps and returns changes, trading more time for more careful reasoning.

For newcomers and experts: If you are new, try the hosted API or a quantized copy first, since this model is far too large for a normal computer. Others have published 148 quantized copies.

If you know the field, the card describes IndexShare, which reuses one indexer across every four sparse attention layers. It says this cuts per-token FLOPs by 2.9 times at 1M context. It also says an improved MTP layer raises acceptance length for speculative decoding by up to 20 percent.

Prompt up to 1M tokens
Sparse attention with shared indexer
Chosen thinking effort
Text or code answer

Title: SWE-bench Pro score
GLM-5.2: 62.1
GLM-5.1: 58.4

Good to know:

• License: MIT
• Cost to run: large GPU clusters for local use, or the Z.ai API Platform, whose price the card does not give
• Data it reaches: the files you download and the prompts you send
• Versions: others have published 148 quantized and 27 fine-tuned copies

On Hugging Face, 13 November 2026

• Task: text generation
• Library: transformers
• Parameters: 753B
• Weights: about 1,507 GB in BF16; about 377 GB at 4-bit (estimate)
• Formats: Safetensors
• Downloads in the last 30 days: 316,074
• Likes: 5,206
• License: mit
• Languages: en, zh
• Quantized versions: 148
• Fine-tunes: 27
• Tags: transformers, safetensors, glm_moe_dsa, text-generation, conversational

Links

• Model: https://huggingface.co/zai-org/GLM-5.2
• Files: https://huggingface.co/zai-org/GLM-5.2/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download zai-org/GLM-5.2

Then follow the usage in its model card. Nothing in this kit runs by itself.