CodeGeeX
ActiveOverview
CodeGeeX is a multilingual code generation model with 13 billion parameters, pre-trained on 850 billion tokens across 23 programming languages. It provides code completion, generation, translation between languages, explanation, and review through a web interface at codegeex.cn and IDE extensions. Targeted at developers, it stands out with support for cross-lingual code translation and open-source availability for research and local deployment.
Key Features
- Multilingual Code Generation - Generates executable code in languages including Python, C++, Java, JavaScript, and Go.
- Crosslingual Code Translation - Translates code snippets between different programming languages with high accuracy.
- Code Completion - Provides autocomplete suggestions based on context in VS Code and JetBrains.
- Code Explanation - Explains code snippets line-by-line using natural language.
- Code Review - Analyzes code changes and provides optimization suggestions.
- Unit Test Generation - Generates unit tests for given code.
- Bug Fixing - Suggests fixes for identified code bugs.
Pricing
| Plan | Price | Includes |
|---|---|---|
| Free | Free | Full access to model, web demo, VS Code/JetBrains extensions, open-source weights. |
Platforms & Requirements
Available as a web app at codegeex.cn, VS Code extension, and JetBrains plugin on Windows, macOS, and Linux. Local inference requires high-end hardware like NVIDIA V100/A100 or Ascend 910 with at least 40GB VRAM for the 13B model. Web version has usage limits for free tier.
Integrations & Ecosystem
- VS Code
- JetBrains IDEs
- Tencent Cloud Studio
- Ascend 910/NVIDIA GPUs
- GitHub (model weights)
- API access
Alternatives
| App | Difference |
|---|---|
| GitHub Copilot | Proprietary, subscription-based with broader context understanding but less focus on multilingual translation. |
| Amazon CodeWhisperer | AWS-integrated, enterprise-focused with security scanning, requires account. |
| Tabnine | Supports local models, privacy-focused, freemium with pro features. |
| Codeium | Free for individuals, supports 70+ languages, emphasizes speed and local deployment. |
Reputation
CodeGeeX is praised for its open-source nature, multilingual capabilities, and free VS Code extension, with user studies showing 83.4% reporting improved coding efficiency. It has tens of thousands of daily active users generating billions of tokens weekly. Criticisms include hardware demands for local use and occasional inaccuracies in complex code generation compared to newer proprietary models.
Sources (8)
- https://keg.cs.tsinghua.edu.cn/codegeex/
- https://github.com/zai-org/CodeGeeX
- https://marketplace.visualstudio.com/items?itemName=aminer.codegeex
- https://realpython.com/ref/ai-coding-tools/codegeex/
- https://arxiv.org/html/2303.17568v2
- https://dorahacks.io/hackathon/codegeex/guide
- https://nl2code.github.io/posts/CodeGeeX/
- https://www.youtube.com/watch?v=bklaDukaCLc