SSM

taxonomy axis: sequence_mixing

Decision rule: State-space layers (Mamba-style) do the mixing; no attention in the block.

Anchor: Mamba

Models in this category (2)

ModelOrgDateBlockAttentionContextTotalActive
xLSTM-7BNXAI (NX-AI)2025-03-17denselinear/state-space8,1926.87B6.87B
Nemotron 3 NanoNVIDIA2025-12-23sparse-MoEGQA524,28831.6B3.2B

What varies within this category

attention variant varies — GQA, linear/state-space. block type varies — dense, sparse-MoE. trained context varies — 524288, 8192. openness varies — open-weights, open-weights-open-data. scale class varies — large, medium.