Alibaba's Qwen grew from a 7B closed chatbot to a 2.4T open model in three years

Alibaba’s Qwen artificial intelligence line has grown from a closed, invite-only chatbot into one of the largest open-weight models released so far, according to an account published October 4, 2026, by the outlet MarkTechPost. Open-weight means the trained files are free to download. The piece traces Qwen from its start in April 2023 through eight major releases, ending with a version that holds 2.4 trillion parameters, the internal numbers a model learns during training.

The first Qwen, MarkTechPost says, ran on a 7-billion-parameter model and was not open source. Access went through an invitation system, so people had to request entry before they could try it. That is a common way for a company to test a product with a small group before opening it to everyone else.

By August 2026, the project had changed on two fronts at once, according to the report. The model had grown enormously by parameter count, and Alibaba had made it open-weight, so the trained files could be downloaded and modified by anyone, rather than kept behind a login or an API that only Alibaba controls.

MarkTechPost frames its piece as a release-by-release account, walking through eight versions of Qwen across roughly three years and noting, release by release, how the license terms shifted alongside the model’s size. The outlet says every factual claim in the story links back to its original source, which suggests the piece works as a reference timeline rather than a single reported story.

The summary available here does not include the names, exact parameter counts, or release dates of the six releases between the first 7-billion-parameter chatbot and the final 2.4-trillion-parameter model. It also does not specify what the license terms looked like at each step, only that they changed over time. Readers who want that level of detail will need the full MarkTechPost account rather than this summary of it.

What is established is the shape of the arc: a company that started by limiting access to a relatively small, closed model ended up giving away one of the largest models in the open-weight category, over roughly three years. Alibaba has not published, in the material reviewed here, an explanation of why it chose to open-source the model rather than keep it closed. What to watch next is whether outside developers and researchers can realistically run or modify the 2.4-trillion-parameter model themselves, since a model that size needs substantial computing hardware no matter what its license says. MarkTechPost’s full release-by-release breakdown, once read in full, should fill in the six steps missing from this summary between the 7-billion and 2.4-trillion-parameter versions.