More Agents, More Samples, or More Thinking? Untangling Coordination, Ensembling, and Test-Time Compute in Multi-Agent LLMs
4.60T1 sourcearXiv cs.MA
Source record
Published by arXiv cs.MA (T1 source). The original is at https://arxiv.org/abs/2611.17610.
Pipeline notes
The summary and note below are generated by the signal pipeline — they are Beyond Desk’s reading, not quotations from the source.
SummaryA synthesis of nine empirical evaluations covering 516 multi-agent configurations finds no overall accuracy gain over baselines matched for samples or compute. Possible gains require partitioned evidence, external verification, or error diversity; the paper includes reviewer checks and accounting utilities.
Why it mattersThe resource-accounting framework separates coordination effects from sample and compute differences, providing a concrete method for testing whether additional agents improve results.
Cited by
No citations on record.