Alibaba transitions Qwen3.6-Max and 3.7-Plus to paid API-only

Alibaba’s Qwen model family experienced explosive growth in 2025, climbing from roughly 1.2% of global self-hosted AI usage to approximately 30% by year's end, overtaking Meta’s Llama in self-hosted deployments. This remarkable rise was fueled by Alibaba’s two-year strategy of distributing Qwen openly and for free, building a massive user base and establishing a genuine platform position. However, the company now faces the classic open-source monetization challenge: self-hosted users already have the model weights and pay nothing for inference. Alibaba’s answer, as reported by The New York Times, is to shift its flagship generations — Qwen3.6-Max and Qwen3.7-Plus — from open source to a closed, API-only model with usage fees attached.

To catalyze adoption of the paid tier, Alibaba has committed a ¥3 billion promotional budget tied to the 2026 Lunar New Year, with cloud infrastructure buildout running through 2027. This playbook mirrors OpenAI and Anthropic: build an audience with accessibility, then monetize through enterprise API contracts. Yet the pivot is not seamless. The closed models must be meaningfully better than the last free versions, and enterprise buyers must be willing to route production traffic through Alibaba Cloud rather than running older Qwen weights on their own hardware. The reporting does not provide a conversion rate from open-source users to paying customers, nor specify how much of the ¥3B targets individual developers versus enterprise sales. The real test will be whether Alibaba’s managed-inference pricing and reliability can undercut the cost of self-hosting, turning its open-source funnel into a durable revenue engine — or risk handing the self-hosted crown back to Llama.