Skip to content
Live
Synced·Added· 12 hours ago35

Models like Kimi K3.1 Pro adopt a hybrid architecture to improve Agent runtime efficiency

Brief overview

Models such as Kimi K3.1 Pro, GLM-5.3-Flash, Qwen3.8-Flash-Next, and Nemotron 3 Super use a hybrid architecture, increasing Agent runtime throughput by 3 times through pre-filling hybrid state cache. These models integrate Kimi Delta Attention (KDA), Gated DeltaNet (GDN), Mamba-2, and other linear attention and state space layers.

Source: Synced · jiqizhixin.com