GLM-130B is a 130 billion parameter open bilingual (English & Chinese) large language model developed by Tsinghua KEG, released in 2022 as one of the earliest accessible 100B+ scale dense LLMs.

Is the Hugging Face demo for GLM-130B working?

No, the space at zai org/GLM-130B currently shows a runtime error due to storage limits exceeded; the original demo was hosted elsewhere, but the model weights remain downloadable from GitHub or Hugging Face repos.

How can I run GLM-130B today?

You can download the model from the official GitHub (zai org/GLM-130B) and run inference on multi GPU setups (e.g., 8x A100 for full precision or 4x RTX 3090 with INT4 quantization) using the provided code.

Is GLM-130B still relevant in 2026?

While groundbreaking in 2022 for its efficiency and openness, newer models like GLM-4 series or Llama 3 offer superior performance; it's mainly of historical/research interest now for bilingual or low resource large model studies.

GLM-130B

Open Bilingual 130B Parameter LLM for English and Chinese Tasks

From Tsinghua KEG (Knowledge Engineering Group)

Founder: Jie Tang

Text Generator

Oct 2022

100K+

Free

Pricing Model

Free

Starting Price

$0/Month

About This AI

GLM-130B is an open-source bidirectional dense language model with 130 billion parameters, pre-trained using the General Language Model (GLM) algorithm.

Developed by Tsinghua University’s KEG lab and collaborators, it excels in bilingual (English & Chinese) understanding and generation, supports efficient inference on limited hardware like a single 8x A100 server or quantized on 4x RTX 3090 GPUs, and performs strongly on downstream NLP tasks with features like low-bit quantization for accessibility.

Chatbot Coding Conversation LLM Text Generator

Pricing

Pricing Model

Free

Starting Price

$0/Month

Key Features

130 billion parameters bidirectional dense model
Bilingual pre training on 400B+ tokens (English & Chinese)
Efficient inference on single 8x A100 or quantized 4x RTX 3090
Supports blank filling and left to right generation
Strong zero shot and few shot performance on diverse NLP tasks

Pros

Fully open source with weights and code available
Exceptional hardware efficiency for a 100B+ model
Balanced strong performance in both English and Chinese
No performance degradation with INT4/INT8 quantization
Pioneering accessible large scale bilingual LLM

Cons

Demo space currently inactive due to storage limits
Requires significant GPU resources even quantized
Older model compared to modern LLMs like Llama 3 or GLM-4
Limited to text based tasks without multimodal support
Potential outdated performance on latest benchmarks

GLM-130B remains a landmark open source release for bilingual large language models, ideal for researchers and developers interested in efficient inference of 100B scale models or historical context in Chinese LLM development.

FAQs

What is GLM-130B?
GLM-130B is a 130 billion parameter open bilingual (English & Chinese) large language model developed by Tsinghua KEG, released in 2022 as one of the earliest accessible 100B+ scale dense LLMs.
Is the Hugging Face demo for GLM-130B working?
No, the space at zai org/GLM-130B currently shows a runtime error due to storage limits exceeded; the original demo was hosted elsewhere, but the model weights remain downloadable from GitHub or Hugging Face repos.
How can I run GLM-130B today?
You can download the model from the official GitHub (zai org/GLM-130B) and run inference on multi GPU setups (e.g., 8x A100 for full precision or 4x RTX 3090 with INT4 quantization) using the provided code.
Is GLM-130B still relevant in 2026?
While groundbreaking in 2022 for its efficiency and openness, newer models like GLM-4 series or Llama 3 offer superior performance; it’s mainly of historical/research interest now for bilingual or low resource large model studies.