# CORAL preprint reports gains from persistent multi-agent search

_Friday, September 4, 2026 at 12:00 AM EDT · Science · Latest · Tier 2 — Notable_

A CORAL research preprint describes a multi-agent framework for open-ended discovery that uses persistent shared memory, asynchronous execution and heartbeat-based interventions rather than fixed exploration rules. The authors report state-of-the-art results on 10 mathematical, algorithmic and systems-optimization tasks, with improvement rates three to 10 times higher than fixed evolutionary-search baselines. On Anthropic's kernel engineering task, four co-evolving agents reportedly improved the best score from 1363 to 1103 cycles. The work has not been independently verified.

## Sources

- [cs.AI updates on arXiv.org](https://arxiv.org/abs/2604.01658)

---
Canonical: https://techandbusiness.org/newswire/_0YRaU3X-l001FsHeiPx9x
Retrieved: 2026-09-04T16:56:19.575Z
Publisher: Tech & Business (techandbusiness.org)