Chinese Frontier Models Shift to Linear Attention and Open Source

Nearly all Chinese frontier models except DeepSeek have adopted linear attention, with most combining sparse attention and Muon for efficiency, a shift that analysts say is paying off alongside China's open-source strategy.

2026-08-26 ~ 2026-08-27 · 2 related posts

• Chinese frontier models adopt linear attention; Zai releases MIT-licensed 320B GLM-5.3-Flash — ivan_bezdomny · 2026-08-26
• Chinese Frontier Models Shift to Linear Attention and Open Source Strategy — SumitGup · 2026-08-27