Skip to content
AI IntelligenceOct 13, 2026AI Intelligence
Article

NVIDIA describes ReplaySSM for faster hybrid-model decoding with no availability stated

NVIDIA describes ReplaySSM, a state-storage design for faster decoding in hybrid models that interleave state-space model layers with attention layers. The announcement names NVIDIA Nemotron 3, Qwen3.5 through Qwen3.8, Kimi-K3, and GLM-5.3-Flash. No release condition or availability is stated.

Frontier EditorialSource: NVIDIA Generative AI
01

Source Brief

NVIDIA describes ReplaySSM for faster hybrid-model decoding with no availability stated: NVIDIA describes ReplaySSM, a state-storage design for faster decoding in hybrid models that interleave state-space model layers with attention layers. The announcement names NVIDIA Nemotron 3, Qwen3.5 through Qwen3.8, Kimi-K3, and GLM-5.3-Flash. No release condition or availability is stated.