Two days after Moonshot AI's Kimi K3 shook the chip market, Alibaba answered with its own reveal: a preview of Qwen3.8 Max, described as a 2.4 trillion-parameter model. Alibaba's US-listed shares jumped more than 3 percent on the news. What most coverage leaves out is the awkward detail underneath it all: Alibaba owns roughly 36 percent of Moonshot AI, so it's effectively answering a move made by a company it partly owns.
A model nobody has verified yet
Qwen3.8 Max first showed up anonymously on LMArena under the codename "Kaleb," where it posted strong front-end results before Alibaba claimed it publicly. The company says the new model beats its predecessor, Qwen3.7-Max, at coding, full-stack development, data analysis, and office workflows, and ranks second overall behind Claude Fable 5. None of that is backed by published results yet: Alibaba hasn't released a technical report, a model card, or a full benchmark table.
A rival Alibaba owns a third of
Moonshot AI released Kimi K3 on July 17, calling it the largest open-weight model in the world at 2.8 trillion parameters. Alibaba fired back with its own reveal just two days later, something investor Michael Burry called a direct response to Moonshot's move. It's rare for a company to compete this openly with a business it holds nearly a third of, and it shows how far the pressure of the AI arms race outweighs ordinary shareholder logic.
What independent testing actually shows
In one head-to-head test run on the same 269-file repository, Kimi K3 scored 83 out of 100 against Qwen3.8 Max's 80. Kimi finished the task faster and used fewer tokens, while Qwen produced cleaner boundaries between system modules and stronger replay metadata for tracking changes. The gap is small and depends on what a given team is optimizing for, and it doesn't back up Alibaba's claim of a clear second-place finish worldwide on its own.





Comments
Discussion
Join the conversation around this story.
Join the discussion
Sign in to comment and reply to other readers.
Sign inNo comments yet
Start the discussion first.