今日速览
| 序号 | 标题 | 来源 | 日期 | 主题 | 推荐等级 |
|---|---|---|---|---|---|
| 1 | ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search | arXiv | 2026-09-11 | AI-Agent | 高 |
| 2 | Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement | arXiv | 2026-09-11 | AI-Agent | 高 |
重点论文与技术动态
1. ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
- 来源：arXiv
- 日期：2026-09-11
- 作者/机构：Jiyan He, Guang Liang, Hao Liu, Haoxiang Guan, Jinbo Sun, Junyi Guo, Wenjun Feng, Yantai Xie, Yifei Shen, Bin Shao, Chuyang Wei, Kai Chen, Kexin Zhou, Minghang Zhu, Shuxin Zheng, Tie-Yan Liu, Taine Zhao, Wenhui Zhu, Xueyin Xu, Xiaoqing Zhang, Yatao Li, Yuxuan Ren
- 主题标签：
AI-Agent,arXiv - 推荐等级：高
- 分类：cs.AI
一句话结论
ZGCM-1 是一个完全开源的 7B 稠密基础模型，通过结合内部推理与外部工具调用，在数学推理和智能体搜索任务上可与参数量大数十倍的前沿模型竞争。
核心内容
- 核心理念：紧凑模型无法被动记忆开放网络，需通过”内部思考+主动工具使用”突破参数容量限制，支持 256K 上下文。
- 架构与系统协同设计：交错门控滑动窗口与全注意力机制，配合 FP8 Muon 优化器；中期训练将交互轨迹重构为 MDP，上下文从 16K 渐进扩展至 256K。
- 在数学推理和 agentic search 基准上，ZGCM-1-7B 可与 Qwen3-235B-A22B、GLM-5.1 等前沿模型竞争；16K 预训练 time-to-loss 效率提升约 4.2 倍。
方法与数据
- 训练流程涵盖预训练、中期训练（MDP 课程）和后训练（SFT），采用 AI 原生研发工作流，由 agent swarms 自主管理集群操作、数据策展与诊断评估。
- 摘要未明确具体训练数据规模与来源细节。
价值判断
- 值得关注：7B 模型在数学推理和 agentic search 上逼近 235B 级模型，对资源受限场景有实际意义。
- 可复用点：交错门控滑动窗口+全注意力的架构设计、MDP 中期训练范式、FP8 Muon 优化器配置。
- 局限/待核查：摘要未提供具体基准分数与消融实验细节，“competitive”的量化程度需查阅正文确认。
摘要
arXiv:2609.13356v1 Announce Type: new Abstract: In this work, we present ZGCM-1, a fully open 7B dense foundation model trained from scratch with extreme data, system, and algorithmic efficiency. ZGCM-1 is founded on a core premise: compact models cannot passively memorize the open web, but can overcome parametric capacity limits by coupling deliberate internal thinking with active external tool use. To support this paradigm across a 256K context, we develop an end-to-end, high-efficiency open training recipe: Architecture & System Co-design: interleaved gated sliding-window and full attention, and a stable FP8 Muon optimizer; Progressive Curriculum & MDP Mid-Training: context scaling across 16K, 64K, and 256K, and the reformulation of interaction traces into Markov Decision Processes. Furthermore, we establish an AI-native R&D workflow where agent swarms autonomously manage cluster operations, data curation, and rapid diagnostic evaluation. Extensive evaluations show that ZGCM-1-7B is competitive across 7B model family on general benchmarks. On several challenging mathematical reasoning and agentic search suites, it remains competitive with frontier models orders of magnitude larger, such as Qwen3-235B-A22B and GLM-5.1. We also show that our pre-training design offers a ~4.2x efficiency improvement in 16K pre-training time-to-loss. Across the full development lifecycle, we distill eight actionable empirical findings-spanning architectural scaling, SFT quality pruning, long-context generalization, and agentic co-training dynamics. To facilitate community research, we open-source model weights from the pre-training, mid-training, and post-training stages, intermediate checkpoints, training code, per-stage data and data recipes, and W&B logs.2. Generalized Agent Iteration: One Formal Framework for Iterative Policy Improvement and Recursive Self-Improvement
- 来源：arXiv
- 日期：2026-09-11
- 作者/机构：Hongyao Tang, Yi Ma, Pengyi Li, Yifu Yuan
- 主题标签：
AI-Agent,arXiv - 推荐等级：高
- 分类：cs.AI
一句话结论
本文提出 Generalized Agent Iteration（GAI）框架，将经典迭代策略改进（GPI）与递归自我改进（RSI）统一为同一学习范式的两个特例，为 RSI 提供了首个基于经典理论的形式化刻画。
核心内容
- 将 agent 定义为系统内可修改组件的配置，学习过程建模为”agent 评估—agent 改进”的循环。
- 两个关键维度：改进机制是否属于 agent 本身（划定 GPI 与 RSI 的边界）；评估标准是否锚定在 agent 外部（决定系统极性为 anchored、goal drift 或 fully self-referential）。
- 利用这两个坐标对现有系统进行统一定位，并逐条件陈述 RSI 的缺陷。
方法与数据
- 摘要未明确具体实验或数据集，本文属于形式化框架类工作。
价值判断
- 值得关注：首次将 RSI 纳入与 GPI 共享的形式化体系，使不同系统可在同一坐标系下比较。
- 可复用点：两个”旋钮”维度可作为分析和设计新 agent 系统的原则性工具。
- 局限/待核查：摘要未给出实证验证，框架的实际适用边界与可操作性有待后续工作检验。