Research·EASYHUB JOURNAL

HySparse2 studies two-level KV sharing for long-context agent inference

Xiaomi MiMoSource published Sep 22, 2026

Editorial illustration accompanying long-context research — Editorial illustration · Not a figure from the paper

What changed

Xiaomi's HySparse2 paper combines two-level KV sharing, token-level sparse selection and a shorter prefill path to reduce the overhead of long tool-rich agent histories. Experiments use an 80B-total, 3B-active MoE model to compare retrieval and inference costs. These are architecture research results, not a release of a complete MiMo-V3 model.

Read original source

Original title: HySparse2: Hybrid Sparse Attention with Two-Level KV Sharing
Source: Xiaomi MiMo · arxiv.org
Topic: Research
Source month: 2026-09

This is a concise EasyHub summary of the linked source, not the full report or original reporting. Availability and preview conditions are described in the summary and original.

Summary page published 2026-09-27 · Editorial information