Qwen Code in-depth review: can it replace Claude as the new development workhorse?

Environment Configuration: Indirect Calls to Qwen Code via Claude Code

A special test path has been chosen for this review: in the Claude Code in the client, by configuring the forwarding address to call the Alibaba-promoted Qwen Code Large models. This was done partly to follow established operating practices, and partly to test the Qwen Code In a real development environment, is it possible to seamlessly replace the Claude 4 Similar tools.

After completing the configuration according to the relevant guidelines, the client's baseurl The address will point to the AliCloud platform, indicating that Qwen Code has been successfully accessed.

Preliminary Testing: Project Initialization and Documentation Generation

To ensure the fairness of the review, this test follows the same pattern as the previous reviews CodyBuddy The exact same steps and cue words are used at the time.

The test starts with a uniapp The initialization of the applet project begins. When the initial version of the project is completed and the first git At the time of submission, the project's preset husky The tool triggers a commit message validation.husky is a commonly used Git Hook tools that can ensure that code submissions follow the specifications set by the team.

In this session.Qwen Code Demonstrated good process understanding. It automatically recognizes that the submission information does not match the conventional commit problems with the specification and made the corrections autonomously. This ability to adapt automation to development workflows is a positive sign.

Next is the initialization of the project documentation. Following the prompt wordQwen Code Generated project documentation containing the basics of the core technology stack, important configuration files, authentication system, style system, etc., which performed moderately well.

The final directory structure of the generated document is as follows:

However, the documentation fails to cover unibest Some of the key features of the framework, such as the file-based routing mechanism and the uni Plug-in system. The testers passed the Claude Code built-in fetch command, which pulls this missing information externally and adds it to the document. This means that the Qwen Code Deeper knowledge of specific frameworks still needs to be strengthened.

Core coding: stability and performance bottlenecks

Once the documentation was ready, the review moved into the core coding phase. The tester entered the same development requirements prompt words as in the previous test.Qwen Code Upon receipt of instructions, overall project development planning begins.

In the early stages, the model was very responsive, outputting 1.5K in just 10 seconds tokensThe

Benefiting from previously supplemented documentation information.Qwen Code The standardized page configuration was correctly applied when generating the page. This proves that the model has good contextual understanding and is able to absorb and utilize new knowledge.

But it didn't last long, and the first problem soon appeared: the connection timed out. The exact cause is unknown, presumably related to the officially deployed computing resource limitations. Interrupting and retrying could solve it temporarily.

Subsequent to the encoding process, the performance drops drastically and even lags for long periods of time. One recording showed that only 9 outputs were made in 108 seconds tokensThe service is largely unavailable.

More confusingly, the system reports "Interrupted by user" without any intervention, and then automatically resumes execution immediately afterward. This erratic behavior creates a great deal of uncertainty in the development experience.

Code Quality: Strange Logic and Instruction Following Issues

In assessing actual coding capabilities, theQwen Code demonstrates some strange programming logic. For example, after modifying the index.vue file, it first renames the original file to index.vue.bakThe file is then modified. This operation directly leads to subsequent processes reporting errors because the original file cannot be found.

Although it was subsequently able to correct this error, the creation of the about page, it again reproduces the same .bak Backup operations. This unconventional and error-prone coding habit is one of its major flaws.

From the output UIQwen Code The design capabilities are mediocre. Without providing specific design elements, it generates a homepage that is neither good nor bad, but the use of gradient colors looks cheap and has obvious signs of AI generation. There is also an icon missing from the page, an omission that the model itself does not notice.

In terms of directive compliance.Qwen Code The performance is equally poor. The prompt explicitly calls for the home page to contain "a rotating image and list of items", but it's designed to be an entry page instead. Worse, it directly removes the item's bottom tab bar, which is a basic component of the applet and is already present in the framework's default configuration.

Cost Analysis and Evaluation Findings

Another important dimension of this test is cost. After creating the about At the time of the page, the free 1 million Tokens The quota was depleted. In order to complete the test, the tester had to clear the context (Clear) three times midway through the test, which suggests that the Qwen Code When dealing with complex development tasks token Consumption is huge.

After the quota was exhausted, the account even showed an outstanding fee of $15. The high fees, combined with the aforementioned stability and intelligence issues, make the current version of the Qwen Code Not yet of a standard that can be used for day-to-day development work.

Thus, the review ends here. The conclusion is clear:Qwen Code There is currently no effective alternative Claude CodeThe

Nevertheless.Qwen Code It shows potential in certain aspects (e.g., understanding the development process). It is foreseeable that as Alibaba subsequently invests more computing resources, optimizes API performance and iterates based on user feedback, theQwen Code There is a strong possibility that in the future it will become Claude A strong contender for the series model.