Alibaba cuts Qwen Code free requests to 100

Alibaba has removed the free OAuth tier for Qwen Code and cut the daily free request quota from 1,000 to 100. The change appeared on the Qwen Code GitHub repository on the day the company updated the project documentation. A repository message read, “Qwen OAuth free tier has been discontinued.”

The repository and accompanying documentation tell developers they can run Qwen Code on Alibaba Cloud’s Coding Plan, through third-party hosts such as OpenRouter or Fireworks AI, or by supplying their own API key. Alibaba lists Coding Plan Pro at $50 per month. The update gives users of the previous free tier an immediate reduction in cloud-based free access and may push some to paid hosting or self-hosted setups.

Qwen Code is a terminal-based coding agent that runs Qwen3-Coder models, supports multi-file repositories and was judged on coding benchmarks such as SWE-Bench. Alibaba has kept core model weights available for local use, but larger Qwen variants require substantial compute to run on local hardware.

Two days before Alibaba’s change, MiniMax altered distribution terms for a major model release. MiniMax pulled M2.7, a 230 billion-parameter model noted for strong coding performance, and replaced an MIT-style license with terms requiring written authorization for commercial use. MiniMax’s Head of Developer Relations wrote that the company made the change to prevent hosting providers from distributing degraded copies under the MiniMax name and added, “They walk away thinking MiniMax is mid.”

Company sources report internal shifts at Alibaba’s AI organization and departures of key leaders as factors behind a move toward more proprietary development. The company did not provide a detailed public explanation in the Qwen Code changelog.

Usage of Chinese open-source models in self-hosted deployments rose from about 1.2% of global usage in late 2024 to roughly 30% by the end of 2025, with Qwen surpassing other models to become widely deployed. Developers and companies that provided free hosting and APIs helped that growth.

Companies restricting access to models have cited concerns about misuse, degraded forks and the need to generate revenue to fund development. U.S. export controls on advanced chips and heightened regulatory scrutiny are factors companies have said complicate distribution and hosting decisions.

Developers now face several options: pay for managed cloud access, route requests through third-party API proxies, run models locally if they have sufficient hardware, or continue with a sharply reduced free quota for cloud requests.