Today, Alibaba announced the discontinuation of the free tier for its Qwen Code service. The notice in the GitHub repository is straightforward: ‘Qwen OAuth free tier has been discontinued.’
In addition to ending the free tier, Alibaba has reduced the daily free request quota from 1,000 to just 100. Users interested in using Qwen code on the cloud are now directed towards alternatives such as Alibaba Cloud Coding Plan, OpenRouter, Fireworks AI, or acquiring their own API key.
Qwen Code was more than a mere hobbyist endeavor; it served as Alibaba’s coding agent for developers, directly competing with Claude Code by utilizing the Qwen3-Coder models. These models support multi-file repositories and boast SWE-Bench scores that rival the best paid tools available in the market.
For those unaware until now, the GitHub changelog missed this update. The Coding Plan Pro subscription is priced at $50 per month.
This decision followed closely on the heels of a similar move by another Chinese AI firm, MiniMax. Just 48 hours prior, MiniMax had withdrawn its M2.7 model—a 230 billion-parameter model nearly matching Claude Opus 4.6 in coding benchmarks—and restructured its licensing to necessitate written permission for commercial use.
The number of parameters is crucial as it defines an AI’s capability to learn, reason, and retain information; more parameters equate to a broader knowledge base.
Initially released under what was termed “MIT-style” terms by MiniMax, the MIT license traditionally does not impose restrictions on commercial applications. However, this change was quickly spotted by developer communities on platforms like Hacker News and Hugging Face within hours.
MiniMax’s Head of Developer Relations cited protection against hosting providers misrepresenting degraded versions as genuine offerings from MiniMax as a reason for the licensing change. Despite these concerns, the new commercial restrictions were maintained.
According to a report by The Financial Times, Alibaba has been steering its Qwen development towards proprietary models following significant leadership changes within the company. Similarly, Xiaomi released MiMo v2 with a closed-source license last month.
Chinese open-source AI models saw their global usage surge from 1.2% at the end of 2024 to about 30% by late 2025, with Qwen surpassing Meta’s Llama as the most widely deployed self-hosted model worldwide.
This growth was largely driven by free services rather than benchmark performance.
With tightening U.S. export controls on chips and ongoing competition in AI between Beijing and Washington, sustaining ‘free’ offerings has become challenging amidst investor demands for returns and scrutiny from the U.S. government over deployment choices.
Those interested in experimenting with Alibaba’s models locally without cost can still do so, as their models are open source. However, utilizing more powerful versions necessitates robust hardware.