March 4, 2026
Capybaras carry the drama
Did Alibaba just kneecap its powerful Qwen AI team?
Top brains walk after ‘tiny but mighty’ drop — fans blame corporate meddling, maybe Elon
TLDR: Qwen’s lead architect and colleagues quit a day after releasing tiny, powerful open‑source models, shocking fans. Comments explode with theories—from Elon poaching to corporate meddling—while developers rave about the tech and worry this could derail Alibaba’s open‑source momentum.
Alibaba’s open‑source AI darling Qwen dropped a “tiny but mighty” model series and then—boom—key leaders walked out 24 hours later. Technical architect Junyang “Justin” Lin signed off with a minimalist “me stepping down. bye my beloved qwen.” on X, and two teammates followed. Cue the internet’s popcorn.gif.
Fans who just celebrated pocket‑sized power—models small enough to run on laptops and phones—are now split between “poached by a billionaire” and “corporate clampdown”. One commenter bets Elon snapped them up after praising Qwen’s “impressive intelligence density.” Another blasts it as “mushroom management,” the classic tale: nerds build magic, suits swoop in. There’s even a spicy rumor that Lin was fired, sparking predictions that Qwen4 will flop and hopes the crew launches a new lab.
Amid the chaos, devs gush: these are the first small models they’d call articulate enough for real use at the edge (read: on your device). The release promised a blueprint for “AI workers” that click buttons and write code—then the capybara art of boxes being carried off felt… oddly prophetic.
The community mood? Equal parts awe at the tech, fury at the timing, and meme‑fueled intrigue. Whether this is a heroic exit or a boardroom bust‑up, open‑source believers are bracing for a plot twist—and checking HN for clues.
Key Points
- •Alibaba’s Qwen team released the open-source Qwen3.5 small model series (0.8B–9B parameters).
- •Within 24 hours of the release, technical lead Junyang “Justin” Lin and colleagues Binyuan Hui and Kaixin Li announced their departures on X.
- •Reasons for the departures were not provided; VentureBeat sought comment from Alibaba.
- •Qwen3.5 uses a Gated DeltaNet hybrid architecture and a 3:1 linear-to-full attention ratio, enabling a 262,000-token context window and edge-friendly performance.
- •Qwen’s model family has surpassed 600 million downloads, and Qwen3.5 drew public praise from Elon Musk for “impressive intelligence density.”