Mamba-3: Improved Sequence Modeling using State Space Principles

ICLROral2026

Authors: Aakash Lahoti, Kevin Li, Berlin Chen, Caitlin Wang, Aviv Bick, J Zico Kolter, Tri Dao, Albert Gu
Affiliation: CMU, Carnegie Mellon University
Venue: ICLR 2026
Track: Oral

TL;DR

Mamba-3, an inference-first SSM that pushes on core SSM principles: improved discretization for better quality, complex dynamics for new capabilities, and MIMO updates for efficient inference.

Opening excerpt from the authors’ abstract. source

Read the paper

OpenReview

Topics

efficient mamba

← All ICLR 2026 Oral papers · Browse the whole archive