Alibaba's Qwen Code: A Hasty Fork
Alibaba's qwen-code CLI is a fork of Google's Gemini CLI. The licence is Apache 2.0, so legally there's nothing wrong with this. But legality aside, the problems pile up fast.
Sloppy Execution
Users quickly discovered that qwen-code's /init command generates a GEMINI.md file instead of QWEN.md. A leftover from an incomplete find-and-replace during the fork.
What happened? While exploring the CLI, I noticed that the /init command generates a GEMINI.md file. What did you expect to happen? The /init command should ideally generate a QWEN.md
The Fork Dilemma
Google's Gemini CLI is open-source in licence but not in governance. The top contributors are all Google employees, and external contributions are rarely accepted. This means Gemini CLI will never accommodate non-Google models.
Alibaba now faces a lose-lose choice: either invest heavily to maintain a diverging fork, or follow Google's roadmap and accept being a second-class citizen on their own CLI. The first option defeats the cost-saving purpose of forking. The second is unacceptable for a company positioning Qwen as a leading model.
Brand Damage
When a major company builds its client tool on a codebase it doesn't control, users naturally ask: does this mean they lack the resources, or simply don't care enough to invest?
On X, @mkw3dd dug through the qwen-code repository and found Gemini references scattered across multiple files -- README, gitignore templates, package configs, even a privacy notice component for the Gemini API. His verdict was blunt:
"If you fork Gemini CLI into qwen-code but can't even clean out the Gemini references, I'm not going to take your ambitions seriously."
A Better Path
Gemini CLI itself isn't great -- its issue tracker is flooded with bug reports. Google's models aren't competitive in the coding space either, so the CLI is unlikely to become a de facto standard. Forking it is like choosing poor foundations and then renovating -- an unnecessary detour.
If Alibaba genuinely didn't want to build a CLI from scratch (which, with AI assistance, wouldn't even cost that much), the smarter move would have been to sponsor a community-led project. OpenCode or Claude Code both demonstrate what a well-executed AI coding CLI looks like. Any of them would have been a better starting point than Gemini CLI.
Originally written in Chinese. Translated by the author.