AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。
4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。
AI News / Research
Back to AI news
AREX: A Recursively Self-Improving Agent for Deep Research
AREX is a series of recursive self-improving (RSI) deep research agents that collect evidence through inner research cycles and audit answers with constrai...
Today AI Intelligence Brief
AREX is a series of recursive self-improving (RSI) deep research agents that collect evidence
through inner research cycles and audit answers with constraints through outer self-improvement cycles, initiating targeted research. The 4B dense model and 122B-A10B MoE model significantly outperform same-scale baselines on benchmarks such as BrowseComp, WideSearch, DeepSearchQA, and HLE, and are competitive with models using more activated parameters.
资讯正文 · AI 整理
Intelligence Assessment
Aioga 编辑摘要
AREX 是一系列面向深度研究的递归自改进智能体，结合内层证据收集与外层逐约束审计，并可根据审计结果启动针对性研究。
背景分析
公开材料将其归入论文研究，来源为 HuggingFace Daily Papers 社区热门论文。研究涉及 4B 密集模型与 122B-A10B MoE 模型。
Aioga 观点
Aioga 判断，AREX 的核心关注点不是单次生成答案，而是把证据收集、约束审计和补充研究组织为递归循环，以改善深度研究过程。
影响与后续
材料称，两种模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，并可与使用更多激活参数的模型竞争。 值得关注后续公开材料是否披露更完整的实验设置、逐项基准结果、审计约束设计与针对性研究机制，以便进一步核验效果和适用边界。
Source and Copyright
The readable text on this page was extracted from the public source and organized with attribution, publication time and the original link. Copyright remains with the original author and publisher.
Ingestion channel: Summary aggregation · Source domain: arxiv.org
Source: HuggingFace Daily Papers（社区热门论文）
Original link: Open original source
Aioga archive: Open intelligence page
Content record: summary-fallback · Updated: 2026-07-23T00:00:00.000Z
API RELAY · DEVELOPER INFRASTRUCTURE
API 中转站
统一接入主流 AI 模型 API，为开发、测试与生产环境提供稳定调用入口。
立即访问 api.w173.com
Scan to open this article