Organizational Instability And Talent Attrition

Issue 63 Edition 2026-03-04 6 min read

General

Sources: 1 • Confidence: Medium • Updated: 2026-04-12 10:22

Key takeaways

Junyang Lin announced on X that he is stepping down from the Qwen team.
A possible trigger for Junyang Lin's resignation was an Alibaba re-org that put a newly hired researcher from Google's Gemini team in charge of Qwen, but this is unconfirmed.
Qwen3.5-397B-A17B is described as an 807GB model and was followed by 122B, 35B, 27B, 9B, 4B, 2B, and 0.8B variants.
The corpus identifies as a key watch item whether departing core Qwen members form a new effort or join another research lab.
As of the emergency meeting, 36Kr-reported accounts said no conclusion had been reached on Junyang Lin's whereabouts or whether he would return.

Junyang Lin announced on X that he is stepping down from the Qwen team.
As of the emergency meeting, 36Kr-reported accounts said no conclusion had been reached on Junyang Lin's whereabouts or whether he would return.
Accounts reported that Junyang Lin later posted on WeChat Moments encouraging Qwen colleagues to continue as planned.
Reporting summarized from 36Kr says Tongyi Lab held an emergency all-hands meeting around 1:00 PM Beijing time on March 4 that was attended by Alibaba Group CEO Wu Yongming.
Several other key Qwen members reportedly resigned the same day, including leaders responsible for Qwen code development, post-training, and contributions to Qwen 3.5/VL/Coder.
36Kr reported that Junyang Lin announced his resignation at 0:11 AM Beijing time on March 4 and that it caused significant internal and industry reaction.

A possible trigger for Junyang Lin's resignation was an Alibaba re-org that put a newly hired researcher from Google's Gemini team in charge of Qwen, but this is unconfirmed.
The corpus infers that Alibaba's CEO attending the emergency all-hands suggests the company may try to retain some departing Qwen talent, while noting the situation remains unsettled.

Qwen3.5-397B-A17B is described as an 807GB model and was followed by 122B, 35B, 27B, 9B, 4B, 2B, and 0.8B variants.
There are positive reports that Qwen 27B and 35B perform well for coding tasks while fitting on 32GB or 64GB Macs.

The corpus identifies as a key watch item whether departing core Qwen members form a new effort or join another research lab.

Did Junyang Lin formally leave the Qwen team (and if so, what role transition occurred), or did he return/retain involvement after March 4?
Was there an Alibaba/Tongyi org change that reassigned Qwen leadership to a newly hired researcher (and who is the current accountable lead)?
Which other Qwen members resigned (names/roles), and did any resignations get reversed following internal discussions?
What concrete outcomes came from the March 4 emergency all-hands (e.g., decisions, leadership appointments, retention measures)?
Do Qwen 27B and 35B achieve the reported coding performance under standardized evaluations, and what are the exact memory/latency characteristics on 32GB and 64GB Macs under specified runtimes and quantization?

Execution risk may be rising for the Qwen roadmap due to reported leadership departure, same-day resignations, and unresolved status, potentially affecting delivery reliability and partner confidence.
An internal re-org could be altering accountability for Qwen, which may shift strategic priorities, decision velocity, and retention outcomes, but causality is unconfirmed.
If mid-sized Qwen3.5 variants are truly deployable on commodity Macs with strong coding performance, adoption by developers could broaden, contingent on standardized evaluation and verified runtime footprints.

Company or team communication that clearly names the current accountable Qwen lead and confirms whether Junyang Lin departed, returned, or transitioned roles.
Verified outcomes from the March 4 emergency all-hands such as retention measures, leadership appointments, or confirmed continuation plans for affected areas like code development and post-training.
Standardized benchmark results and reproducible runtime measurements for Qwen 27B and 35B on 32GB and 64GB Macs under specified runtimes and quantization, aligning with reported claims.

Confirmation that Junyang Lin and other reported resignations were reversed or that team continuity was restored with stable leadership, reducing the implied execution-risk delta.
Clear evidence that no meaningful org change occurred affecting Qwen leadership accountability, undermining the re-org as a key explanatory factor.
Independent evaluations showing Qwen 27B and 35B do not meet reported coding performance or cannot achieve the claimed memory and latency characteristics on target Macs.