Neura Market
CodeQwen1.5-7B logo

CodeQwen1.5-7B

Free

Open-source code LLM with 92 languages and 64K context

FreeFree tier
Type
Open Source
Company
Qwen Team (Alibaba Cloud)
Visit Website GitHub (38 stars)

About CodeQwen1.5-7B

CodeQwen1.5-7B is a specialized open-source code language model built upon Qwen1.5, developed by the Qwen Team at Alibaba Cloud. It has been pretrained on approximately 3 trillion tokens of code-related data and supports 92 programming languages. The model features a long-context window of 64K tokens, enabling exceptional understanding and generation of lengthy codebases. In evaluations, CodeQwen1.5-7B demonstrates strong performance on benchmarks like HumanEval, MBPP, LiveCodeBench, and MultiPL-E, often surpassing larger open-source models and narrowing the gap with proprietary models like GPT-3.5 and GPT-4 in code generation, editing, and SQL tasks. It is available under an open-source license, offering a transparent and accessible alternative to proprietary coding assistants.

Key Features

Pretrained on approximately 3 trillion tokens of code-related data
Supports 92 programming languages
64,000-token context window for long-context understanding
Strong performance on HumanEval (83.5% 0-shot), MBPP, LiveCodeBench, and MultiPL-E benchmarks
Capable of code generation, editing, long-context modeling, and SQL tasks
Open-source model, freely available for use and modification

Pros & Cons

Pros
  • Open-source and free to use, providing a transparent alternative to proprietary code assistants
  • Competitive performance on coding benchmarks despite only 7B parameters
  • Broad language support covering 92 programming languages
  • Long 64K token context window for handling large codebases
  • Backed by the Qwen Team (Alibaba Cloud) with active development and community support
Cons
  • Still lags behind top proprietary models (e.g., GPT-4) on some complex benchmarks
  • Relatively small 7B parameter size may limit performance on very intricate tasks
  • Potential data contamination from pretraining on LeetCode challenges (noted in LiveCodeBench evaluation)

Best For

Assisting software developers with code generation and completionCode editing and refactoring across multiple languagesUnderstanding and generating long code files or repositoriesTranslating natural language instructions into executable codeSQL query generation and database interactionEducational tool for learning programming concepts

FAQ

What is CodeQwen1.5-7B?
CodeQwen1.5-7B is a specialized open-source code language model from the Qwen Team, built on Qwen1.5 and pretrained on ~3 trillion code tokens. It supports 92 programming languages and has a 64K token context window.
How does CodeQwen1.5-7B perform on coding benchmarks?
It achieves strong results: 83.5% on HumanEval 0-shot, 78.7% on HumanEval+, 77.7% on MBPP 0-shot, and 70.6% on MBPP 3-shot (Chat model). It also ranks among top open-access models on LiveCodeBench and excels across eight mainstream languages in MultiPL-E.
Is CodeQwen1.5-7B free to use?
Yes, it is open-source and free to use. The model weights and code are available on GitHub, Hugging Face, and ModelScope.
What programming languages does CodeQwen1.5-7B support?
It supports 92 programming languages, including Python, C++, Java, PHP, TypeScript, C#, Bash, and JavaScript.
Can CodeQwen1.5-7B handle long code files?
Yes, it has a context length of 64K tokens, enabling it to process and generate long code sequences and entire codebases.