CodeGeeX2 is an open source multilingual AI model specialized in code generation, built on ChatGLM2-6B and trained on massive code data for tasks like writing, completing, translating, and explaining code in over 100 languages.

Is CodeGeeX2 free to use?

Yes, the model weights are free for academic research and personal use, with full open-source code under Apache 2.0. Commercial applications require filling a registration form.

How does CodeGeeX2 perform compared to other models?

It achieves significant improvements on HumanEval-X (e.g., 35.9% Pass@1 on Python), often surpassing larger models like StarCoder-15B despite having only 6 billion parameters.

Can I run CodeGeeX2 locally?

Yes, it supports efficient local inference especially the INT4 quantized version needs just 6GB GPU memory and integrates with plugins for VS Code, JetBrains, and more.

CodeGeeX2

Powerful Multilingual AI Code Generation Model for Developers

From THUDM / Z.ai

Founder: THUDM (Tsinghua University)

Code & Development

Jul 2023

50K+

Free

Pricing Model

Free

Starting Price

$0/Month

About This AI

CodeGeeX2 is the second-generation multilingual code generation model from THUDM (Tsinghua University), built on the ChatGLM2-6B architecture and further pre-trained on 600 billion code tokens.

It excels at code generation, completion, translation, explanation, debugging, comment generation, and more across over 100 programming languages.

With strong bilingual (Chinese/English) support, 8192-token context, and efficient quantized inference (as low as 6GB GPU memory), it’s designed for both research and practical development use.

Coding Debugging LLM Text To Code

Pricing

Pricing Model

Free

Starting Price

$0/Month

Key Features

Multilingual code generation and completion in 100+ languages
Code translation, explanation, debugging, and comment generation
Supports Chinese and English prompts with 8192-token context
Quantized INT4 version for low memory inference (6GB GPU)
Integration with IDE plugins like VS Code and JetBrains

Pros

Exceptional performance on HumanEval-X benchmarks (+107% over predecessor)
Outperforms larger models like StarCoder-15B in several languages with only 6B parameters
Fully open weights for research and efficient local deployment
Strong bilingual support and fast inference speeds
Comprehensive AI coding assistant features via plugins

Cons

Commercial use requires registration and approval
Requires GPU for optimal performance
No hosted API or cloud service included
Limited to offline/local use without additional setup

CodeGeeX2 is excellent for developers and researchers needing a lightweight, high-performing multilingual code AI that runs locally and integrates seamlessly into IDEs for faster coding workflows.

FAQs

What is CodeGeeX2?
CodeGeeX2 is an open source multilingual AI model specialized in code generation, built on ChatGLM2-6B and trained on massive code data for tasks like writing, completing, translating, and explaining code in over 100 languages.
Is CodeGeeX2 free to use?
Yes, the model weights are free for academic research and personal use, with full open-source code under Apache 2.0. Commercial applications require filling a registration form.
How does CodeGeeX2 perform compared to other models?
It achieves significant improvements on HumanEval-X (e.g., 35.9% Pass@1 on Python), often surpassing larger models like StarCoder-15B despite having only 6 billion parameters.
Can I run CodeGeeX2 locally?
Yes, it supports efficient local inference especially the INT4 quantized version needs just 6GB GPU memory and integrates with plugins for VS Code, JetBrains, and more.