The Story of Qwen: Alibaba’s AI Models From 7B to 2.4T
MarkTechPost Asif Razzaq
Alibaba’s Qwen went from invite-only chatbot to 2.4T open weights. The twist: it’s now split between permissive open models and paid frontier tiers.
Based on reporting by MarkTechPost, Asif Razzaq — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Alibaba’s Qwen story starts with a chatbot demo in April 2023 and ends, so far, with a 2.4 trillion-parameter model released as open weights in August 2026. That arc is fast even by AI standards. It also shows something bigger than raw scale: Alibaba kept changing the rules around access, commercial use, and where the money sits.
The early moves were cautious, then suddenly not. Tongyi Qianwen began as an invite-only enterprise beta on April 7, 2023, before being shown publicly four days later. By August, Alibaba had open-sourced Qwen-7B and Qwen-7B-Chat, with a 2,048-token context and a license that allowed commercial use below 100 million monthly users. Vision support arrived soon after with Qwen-VL, and by December the family had stretched from 1.8B to 72B parameters.
2024 is where Qwen turned into a developer habit. Alibaba shipped Qwen1.5 in February, Qwen2 in June, and Qwen2.5 in September, then layered on coding, audio, reasoning and video models. The licensing picture softened too: by Qwen2, most sizes had moved to Apache 2.0. Qwen2.5 pushed the line further, with Alibaba saying the family had passed 40 million downloads and inspired more than 50,000 derivative models on Hugging Face.
Then 2025 brought the real split. Qwen3 shipped under Apache 2.0 with hybrid thinking, 36 trillion training tokens and support for 119 languages and dialects. But later releases started dividing the lineup into open mid-tier models and closed flagship systems. Qwen3-Max-Preview was API-only, Qwen3.6-Plus and Qwen3.7-Plus were proprietary, and Qwen3.8-2.4T-A95B came with a new commercial threshold: providers earning over US$50 million in 12 months need a license. That is not open source as ideology. That is open source as funnel.
By 2026, the ecosystem had grown huge enough to make the split make sense on Alibaba’s terms. Reuters put Qwen’s downloads at over 1 billion and its open model count above 400. Alibaba also said the Qwen app had 203 million monthly active users in February 2026. The pattern is pretty clear: keep the smaller and mid-sized models permissive, keep developers building, and reserve the frontier for the cloud bill.
My take — AI-written commentary, not fact-checked reporting
Alibaba has figured out the neatest trick in AI: make the base models generous enough to win the ecosystem, then put a toll booth on the shiny stuff. Open weights are great until the invoice arrives. This is how “open” keeps surviving in the real world — with just enough of a catch to pay for the next monster model.
Read more about this at: MarkTechPost