Neura Market
Qwen2.5-Coder-1.5B|7B|32B logo

Qwen2.5-Coder-1.5B|7B|32B

Free

Code More, Learn More!

FreeFree tier
Inputs: text, codeOutputs: text, code
Type
Open Source
Company
Alibaba (Qwen Team)
Visit Website GitHub (38 stars)

About Qwen2.5-Coder-1.5B|7B|32B

Qwen2.5-Coder is an open-source code language model from the Qwen team (Alibaba), available in 1.5B, 7B, and 32B parameter versions (32B coming soon). It is built on Qwen2.5 and trained on 5.5 trillion tokens of code data (source code, text-code grounding, and synthetic data) while retaining strong math and general capabilities. The model supports up to 128K tokens of context and covers 92 programming languages. The base version excels in code generation, multi-programming code generation, code completion, and code repair; the instruction-tuned version (Qwen2.5-Coder-Instruct) further improves performance on code reasoning, math reasoning, and general tasks. Notably, the 7B version outperforms larger proprietary models like DeepSeek-Coder-V2-Lite and CodeStral-22B on code benchmarks, while maintaining competitive math (GSM8K, MATH) and general (MMLU, ARC) performance.

Key Features

Available in 1.5B, 7B, and 32B parameter sizes (32B coming soon)
Trained on 5.5 trillion tokens of code data
Supports up to 128K tokens of context
Covers 92 programming languages
Achieves state-of-the-art results in code generation, completion, repair, and multi-language tasks
Instruction-tuned variant (Qwen2.5-Coder-Instruct) improves code reasoning, math reasoning, and following complex instructions
Retains strong math (GSM8K, MATH) and general (MMLU, ARC) capabilities from Qwen2.5
Open-source and freely available on GitHub, Hugging Face, and ModelScope

Pros & Cons

Pros
  • Open-source and free to use
  • 7B model outperforms larger models like DeepSeek-Coder-V2-Lite and CodeStral-22B
  • Supports very long context (128K tokens)
  • Broad language support (92 programming languages)
  • Maintains strong math and general abilities alongside coding
  • Active community and availability on multiple platforms (Hugging Face, GitHub, ModelScope)
Cons
  • 32B version not yet released (coming soon)
  • May require significant computational resources for larger models
  • Still a research model, not a polished commercial product

Best For

Code generation and completionCode repair and debuggingMulti-language programming (40+ languages in evaluations)Code reasoning and analysisMath reasoning and problem solvingBuilding code agents and automation toolsGeneral assistant tasks (math, logic, instruction following)

FAQ

What is Qwen2.5-Coder?
Qwen2.5-Coder is an open-source code language model developed by the Qwen team (Alibaba), available in 1.5B, 7B, and 32B parameter sizes. It is trained on 5.5 trillion tokens of code data and supports 92 programming languages with a 128K token context window.
How does Qwen2.5-Coder compare to other code models?
The 7B version of Qwen2.5-Coder outperforms larger models such as DeepSeek-Coder-V2-Lite and CodeStral-22B on code benchmarks. The instruction-tuned variant also excels in code reasoning (CRUXEval) and math reasoning (GSM8K, MATH).
What are the model sizes available?
Qwen2.5-Coder is available in 1.5B, 7B, and 32B parameter sizes. The 32B version is listed as 'coming soon'.
Is Qwen2.5-Coder free to use?
Yes, Qwen2.5-Coder is open-source and free to use. It is available on GitHub, Hugging Face, and ModelScope.
What tasks can Qwen2.5-Coder perform?
It can handle code generation, completion, repair, multi-language programming, code reasoning, math reasoning, and general instruction-following tasks. It is also suitable for building code agents.