Infoblock alpha
cs-ai/agents IB-arxiv-2607.28527-manta-avg-score measured

Across five benchmarks (information seeking, tool use, planning, workflow execution, mathematical reasoning), MANTA achieved the highest average score of 74.0, outperforming the strongest baseline by 5.8 percentage points.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

average_score 74.0 · margin_over_best_baseline_pp 5.8 · benchmarks 5

01

Primary source

  1. arxiv · 2607.28527 · v1 · abstract
Title
MANTA: Multi-Agent Network Topology Adaptation for Self-Evolving Multi-Agent Systems
Authors
Mao-xun Huang, Jerry Wang, Yi-Cheng Lai, Zhengxin Zhang, Claire Cardie, Hen-Hsen Huang
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
on five benchmarks spanning information seeking, tool use, planning, workflow execution, and mathematical reasoning. MANTA achieves the highest average score of 74.0, outperforming the strongest baseline by 5.8 percentage points

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.