Something is afoot in the land of Qwen
A key AI research team, including its lead, abruptly resigned from Alibaba's Qwen project, reportedly due to an internal reorg. This surprising exodus casts a shadow over the future of the highly-praised Qwen 3.5 open-source models, which have been lauded for their performance, especially on local hardware. The Hacker News community is abuzz with speculation about the team's next move, the models' real-world utility, and the broader geopolitical implications for AI talent.
The Lowdown
Simon Willison reports on a significant upheaval within Alibaba's Qwen AI team, with lead researcher Junyang Lin and several other core contributors announcing their resignations. This sudden departure, following an internal reorg, has cast uncertainty over the future of the Qwen 3.5 open-source models, which have recently garnered considerable praise.
- Junyang Lin, lead researcher and key figure in Qwen's open-source efforts, publicly announced his resignation. Other core team members, specializing in areas like code development and post-training research, also departed.
- The catalyst appears to be an internal Alibaba reorganization where a new researcher from Google's Gemini team was placed in charge of Qwen.
- Alibaba's CEO held an emergency meeting, indicating the company's awareness of the gravity of the situation and a potential effort to retain talent.
- The recently released Qwen 3.5 models are recognized as "exceptionally good," particularly for their performance at smaller sizes (e.g., 27B, 35B) and suitability for local deployment.
- The author expresses concern that the disbanding of this innovative team would be a "tragedy" given their track record in achieving high-quality results from smaller models.
The situation remains fluid, with the possibility of the team regrouping or starting anew. This unexpected event raises questions about talent management in large tech companies and the future trajectory of a highly impactful open-source AI project.
The Gossip
Qwen's Quandary: Continuity Concerns
Commenters largely expressed disappointment and hope for the Qwen team's future, fearing the departures would hinder development of the "exceptionally good" Qwen 3.5 models. Many wished the team would continue their work, either by starting a new venture or being retained by Alibaba with a "blank check." There's concern that any future models from this team might become closed-source.
Performance Perceptions and Practical Plight
The discussion saw mixed reviews on Qwen 3.5's real-world utility, especially for local coding agents. While some users reported impressive results on Macs for tasks like Rust/Elixir coding and documentation generation, others found its performance "pretty bad" or prone to infinite loops, suggesting it might be overfit to benchmarks. There was also technical clarification on MoE models and the meaning of 'A3B'.
Geopolitics Guiding Genius
Hacker News users debated the broader implications of these resignations for global AI talent. Questions arose about why US labs aren't aggressively recruiting such researchers, with speculation pointing to factors like China's competitive offers, nationalism, quality of life, and perceived US immigration complexities. Some comments veered into sensitive territory regarding US policies impacting Chinese nationals.