A closer look

Hugging Face counted more than 2 billion Qwen downloads in the first seven months of 2026. In the same year, Alibaba's Qwen team lost its technical leader, its post training chief and a senior researcher.

On 4 November, Qwen went into beta inside Apple Intelligence in China. The road there ran through an internal fight over computing power.

36Kr reported tensions over a reorganisation, hiring and compute. At an emergency all hands, researchers asked why paying Alibaba Cloud customers got capacity while the Qwen team stayed short.

The 2026 timeline:
- Early 2026: senior researcher Binyuan Hui leaves
- March: technical leader Junyang Lin exits, and post training chief Bowen Yu also leaves
- June: Token Foundry puts key model teams under CEO Eddie Wu
- August: Apple has a China specific model trained with Alibaba's support
- September: Dayiheng Liu becomes head of the Qwen model project
- 4 November: Apple Intelligence beta in China, with Qwen

Lin's goodbye was three words: "Bye my beloved Qwen."

Management acknowledged communication problems and said the changes were meant to strengthen Qwen.

For founders, this is the frontier lab problem in one story. Models need centralised capital and huge compute. Research teams run on autonomy and speed. The hard question is who gets the GPUs first: the paying customer or the team building the product.

Qwen kept winning users while Alibaba kept losing the people who built it. Apple chose it anyway.