Qwen Code

Watching its progress

Developer

Alibaba Cloud

Version Reviewed

2.5.0

Last Updated

June 2026

Platform

macOS, Linux, Windows

Overview

I started testing Qwen Code in mid-2025. It is Alibaba Cloud's command-line AI coding agent, built on the Qwen model family. It provides code generation, file editing, and terminal integration. It is one of the newer entrants in the CLI agent space.

The model quality has improved noticeably with each release. The CLI tooling is still catching up to the model capabilities.

Personal Note

The model quality improved faster than the CLI tooling, which is unusual in this space.

What Works Well

• Competitive performance on coding benchmarks, particularly for Python and Java.
• Very low cost compared to Western alternatives.
• Good support for Chinese-language documentation and comments.
• Open-weight models available for self-hosting.
• Rapid iteration cycle with frequent model updates.

Where It Works Less Well

• CLI tooling is still immature with rough edges in installation and configuration.
• Documentation is primarily in Chinese. English translations lag behind.
• Smaller community and fewer third-party resources.
• Code quality drops outside of well-represented languages.
• Privacy considerations for code sent to Alibaba Cloud servers.

Use Cases

Developers in the Alibaba Cloud ecosystem or those needing a low-cost coding agent. The open-weight models are interesting for teams that want to self-host.

Engineering Maturity

Low-medium. Handles straightforward tasks in core languages competently. Struggles with complex multi-file changes or less common frameworks.

Product Maturity

Low. The CLI is functional but has usability issues. The documentation gap is significant for English-speaking users.

Developer Experience

Installation and setup are more complex than alternatives. The interface is functional but less polished.

Workflow Integration

Standard terminal integration with filesystem access. The setup friction makes it less seamless than more mature alternatives.

Performance

Model performance is competitive for supported languages. CLI response times are acceptable.

Documentation

Primarily in Chinese. English documentation is available but less comprehensive.

Pricing

Very competitive. Significantly lower than Claude or GPT-4. Open-weight models can be self-hosted at no API cost.

Platform Support

macOS, Linux, Windows. Cross-platform through standard tooling.

Verdict

Qwen Code shows rapid progress, particularly in model quality and pricing. It is not yet ready for my daily production use. I am watching its development, especially as the CLI tooling matures.

Changelog