GLM-5.3-Flash vs Qwen3.8-Flash-Next: Two Chinese AI Labs Independently Converge on the Same Model Architecture

Two independent Chinese AI research labs released fast, efficient language models within a day of each other that use nearly identical internal designs, despite developing them separately. Both models use a hybrid approach combining cheaper linear attention layers with selective full attention, compress context before processing it, replace the traditional single data pathway with four parallel gated branches, and train with the same optimizer and matrix-splitting technique. This convergence suggests the field is settling on shared architectural principles for building efficient models, though the labs disagreed on one component: how to encode position information in the attention layers, with one approach later showing problems after additional training stages that weren't visible during initial development.