Agent

更复杂的SWE-Agent在SWE-pro bench上相比于mini-swe-agent表现更差且出现了实例卡死

研究者每日需面对数十篇新论文，真正耗费精力的往往不是阅读，而是筛选。现有工具大多聚焦于单篇精读，缺乏筛选与分类能力；而许多看似有价值的工作，又常因代码仓库缺失或不完整而难以复现，带来高昂的沉没成本。paper_reader 正是针对这一痛点的自动化工具：它以可复现性为核心筛选标准，结合规则初筛与基于 Kimi CLI 的 Agent Skill 语义判断，每日自动抓取 HuggingFace Daily Papers，完成分类打标签与中文摘要生成，并借助 macOS launchd 定时调度与纯静态页面实现无人值守的更新与展示。本文记录其设计动机、关键取舍与工程实现。

We are open sourcing our latest model, Kimi K2.6, featuring state-of-the-art coding, long-horizon execution, and agent swarm …

We introduce Kimi K2.5, an open-source multimodal agentic model designed to advance general agentic intelligence. The model focuses on …

Interactive analysis platform for ultra-long agent trajectories (400+ turns, 1M+ tokens). Decomposes massive coding agent trajectories into hierarchical step DAGs, scores trajectory quality, and renders interactive visualizations.

We introduce Kimi K2, a Mixture-of-Experts large language model with 32 billion activated parameters and 1 trillion total parameters. …

MIX-MCP is a Go-based aggregator of MCP services aiming to build a foundational layer for interaction between large language models and Web3 technologies.

作者提出CoT-Influx方法，一种对CoT的示例和内容进行优化从而提高LLMs推理能力的方法，其核心思想是通过剪枝最大化有效信息的输入。

3月最新用于游戏的大模型Agent综述