Qwen Code Joins the Pay-to-Play Club: Alibaba Slams the Door on Free Tier Users

Alibaba's Qwen Code officially killed its free tier today, because nothing says "innovation" quite like charging developers for the privilege of debugging your code. The message in the GitHub repo doesn't mince words: "Qwen OAuth free tier has been discontinued"—straight to the point, no fluff, just cold hard paywalls. The company also reduced the 1,000 free requests quota to 100 per day, which is basically a participation trophy at this point. Users wanting to run Qwen code on the cloud are now directed to check Alibaba Cloud Coding Plan, OpenRouter, Fireworks AI, or bring their own API key. Translation: go fund someone else's compute bills.

Qwen Code wasn't some hobby project gathering dust in a repo somewhere. It was Alibaba's terminal coding agent—a direct Claude Code rival running Qwen3-Coder models, with multi-file repository support and SWE-Bench scores competitive with the best paid tools on the market. This was the real deal, the 800-pound gorilla in the room that actually worked. The Coding Plan Pro subscription runs $50 a month, which is apparently the going rate for premium gatekeeping in 2025.

This landed 48 hours after fellow Chinese AI company MiniMax pulled almost the same move, because when one Chinese lab starts locking things down, apparently it's contagious. The Chinese lab dropped M2.7—a 230 billion-parameter model that nearly matches Claude Opus 4.6 on coding benchmarks—then immediately rewrote the license to require written authorization for commercial use. The model launched under what MiniMax called "MIT-style" terms, but the commercial restriction stayed anyway after the developer community on Hacker News and Hugging Face noticed within hours. MiniMax said it was protecting against bad-faith hosting providers shipping degraded versions under its name. Sure, Jan.

Neither move is accidental. The Financial Times reported that Alibaba's own Qwen team has been moving toward proprietary development after key leadership departures—because nothing says "we're pivoting" like watching your best people walk out the door. Xiaomi, another Chinese company, shipped MiMo v2 last month under a closed-source license. The writing's on the wall, and it's written in yuan.

Chinese open-source models went from 1.2% of global open-model usage in late 2024 to roughly 30% by end of 2025. Qwen overtook Meta's Llama as the most deployed self-hosted model on the planet. That adoption wasn't built on benchmarks—it was built on free services, the same way a degen builds a following: give them something for nothing, then slowly start charging. With U.S. chip export controls tightening and the Beijing-Washington AI race grinding on, "free" is harder to defend when investors want returns and the U.S. government is watching every deployment decision like a hawk with a spreadsheet.

Those wanting to play with Alibaba's models locally, and free, still can. Their models remain open source, but the more powerful ones require pretty heavy hardware to run. So yes, you can run Qwen for free—just make sure your GPU rig is beefy enough to mine those sweet, sweet inference tokens. The cloud is closed, but your basement awaits.