每日简报

2026-10-04

← 历史归档

DietrichGebert/ponytail

JavaScript · ★ 154,035 · 🍴 8,285 · 📈 1,281 stars today

Makes your AI agent think like the laziest senior dev in the room. The best code is the code you never wrote.

pbakaus/impeccable

JavaScript · ★ 75,705 · 🍴 4,531 · 📈 699 stars today

The design language that makes your AI harness better at design.

affaan-m/ECC

JavaScript · ★ 272,524 · 🍴 40,689 · 📈 897 stars today

The agent harness performance optimization system. Skills, instincts, memory, security, and research-first development for Claude Code, Codex, Opencode, Cursor and beyond.

Effect-TS/effect

TypeScript · ★ 16,907 · 🍴 818 · 📈 302 stars today

Build production-ready applications in TypeScript

JuliusBrussee/caveman

Go · ★ 109,676 · 🍴 6,344 · 📈 507 stars today

🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talking like a caveman.

Panniantong/Agent-Reach

Python · ★ 90,206 · 🍴 7,932 · 📈 1,696 stars today

Give your AI agent eyes to see the entire internet. Read & search Twitter, Reddit, YouTube, GitHub, Bilibili, XiaoHongShu — one CLI, zero API fees.

pingdotgg/t3code

TypeScript · ★ 24,878 · 🍴 6,471 · 📈 252 stars today

thedotmack/claude-mem

TypeScript · ★ 95,802 · 🍴 8,469 · 📈 79 stars today

Persistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + More

cloudflare/cloudflare-os

TypeScript · ★ 10,691 · 🍴 1,290 · 📈 85 stars today

Agent workspace built on Cloudflare Workers for creating documents, building apps, and running agents with your company’s context and systems.

addyosmani/agent-skills

JavaScript · ★ 100,953 · 🍴 10,606 · 📈 252 stars today

Production-grade engineering skills for AI coding agents.

obra/superpowers

Shell · ★ 295,055 · 🍴 26,370 · 📈 577 stars today

An agentic skills framework & software development methodology that works.

mattpocock/skills

Shell · ★ 275,625 · 🍴 23,121 · 📈 751 stars today

Skills for Real Engineers. Straight from my .agents directory.

mksglu/context-mode

TypeScript · ★ 25,330 · 🍴 1,812 · 📈 256 stars today

Context window optimization for AI coding agents. Sandboxes tool output (98% reduction), persists session memory, and enforces routing across 17 platforms via MCP + hooks.

earendil-works/pi

TypeScript · ★ 112,300 · 🍴 14,239 · 📈 408 stars today

AI agent toolkit: unified LLM API, agent loop, TUI, coding agent CLI

getsentry/sentry

Python · ★ 45,255 · 🍴 4,899 · 📈 214 stars today

Developer-first error tracking and performance monitoring

anthropics/claude-code

TypeScript · ★ 149,339 · 🍴 25,437 · 📈 128 stars today

Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.

OpenCut-app/OpenCut

TypeScript · ★ 91,793 · 🍴 9,068 · 📈 232 stars today

The open-source CapCut alternative

OneStreamer: Unifying Perception, Memory, and Proactive Response in Streaming Video Interaction

👍 162

Streaming video LLMs must retain evidence before its relevance to future tasks is known and respond when sufficient evidence becomes available. The challenge is to form reusable factual memory without compromising real-time perception. We introduce OneStreamer, which jointly learns query-independent

中文介绍 针对流式视频 LLM 需边感知边记忆、并在证据充足时主动响应的难题,提出 OneStreamer。该模型联合学习查询无关的可复用事实记忆与实时感知能力,避免记忆构建损害实时性。目标是让视频模型在流式交互中形成可复用记忆、及时决策,而不破坏即时理解。

On-Policy or Off-Policy Learning? A Systematic Study of Distillation Dynamics

👍 154

On-policy learning has been argued to reduce catastrophic forgetting, produce sparser parameter updates, and improve generalisation. However, existing comparisons between supervised fine-tuning and reinforcement learning vary many factors simultaneously, making the contribution of rollout policy dif

中文介绍 系统研究蒸馏学习中 on-policy 与 off-policy 的差异。已有比较常同时改变多因素,难以归因 rollout 策略的作用。本文控制变量分析蒸馏动力学,检验 on-policy 是否真能减少灾难性遗忘、产生更稀疏的参数更新并提升泛化能力,厘清各因素贡献。

E-MoE: Enhanced Mixture-of-Experts for Non-Factorized Diffusion Language Models

👍 57

Masked diffusion models (MDMs) generate sequences by progressively unmasking several tokens per denoising step, but their reverse process is typically factorized over positions, limiting sample quality in the few-step regime where diffusion's speed advantage over autoregressive decoding matters most

中文介绍 针对掩码扩散模型(MDM)逆过程按位置独立分解、导致少步生成质量受限的问题,提出 E-MoE。该方法引入增强的专家混合机制实现非分解式生成,在扩散相比自回归最具速度优势的少步区域提升样本质量。

Make Sparse Rewards Count: Density-Aware Reward Aggregation for Multi-Reward RL

👍 50

Multi-reward reinforcement learning trains large language models to satisfy multiple behavioral objectives simultaneously. Reward-wise normalization, as used in GDPO, preserves reward-specific relative information within rollout groups, but different objectives can still exhibit uneven learning prog

中文介绍 面向多奖励强化学习,GDPO 的按奖励归一化虽保留组内相对信息,但不同目标学习进度仍不均。本文提出密度感知的奖励聚合方法 Make Sparse Rewards Count,使稀疏奖励更有效贡献,平衡多目标的学习进度,提升同时满足多行为目标的效果。

EgoTools: Towards Tool-Centric Reasoning in Real-World Egocentric Videos

👍 45

Real-world embodied tasks, from everyday activities to professional procedures, require agents to act under physical constraints while tracking evolving object and task states. Tool use sits at the heart of such tasks, as many everyday and professional activities are tool-mediated. Understanding the

中文介绍 面向真实世界中工具中介的具身任务,提出 EgoTools 数据集与基准。第一人称视频中智能体需在物理约束下跟踪不断变化的物体与任务状态,工具使用是核心。该工作推动以工具为中心的推理能力评估,覆盖日常与专业操作场景。

InterEvolve: Test-Time Evolution of Reward Programs for Humanoid Loco-Manipulation

👍 43

We study test-time evolution for humanoid loco-manipulation: solving tasks that a controller was never trained for by repurposing its existing skills, improving from its own attempts, and retaining what it learns, without retraining. Our key insight is that a broad controller already holds much of t

中文介绍 研究人形机器人 loco-manipulation 的测试时演化(InterEvolve)。核心思路是已训练的宽泛控制器已具备大量技能,可通过进化奖励程序复现并改进自身尝试来求解未训练任务,无需重新训练。方法保留学到的知识并持续自我提升。

X-Tree: Tokenizing Reusable Experience for Efficient Agent Generalization

👍 41

Multi-step agents are trained on flat action streams: SFT and RLVR weight every token uniformly and ignore the sub-procedures that recur across tasks, the hierarchy that lets humans plan top-down from reusable routines. This structure sits unused, and flat training uses each scarce trajectory less f

中文介绍 现有多步智能体在扁平动作流上训练,SFT 与 RLVR 对每个 token 等权,忽视了跨任务复用的子程序层级。X-Tree 将可复用经验分块为可复用树结构,让智能体像人类一样自上而下规划,提高每条稀缺轨迹的利用效率与泛化能力。

Decentralized Master-Mind: Joint Action Refinement through Iterative Intent Denoising in Multi-Agent Pathfinding

👍 41

Decentralized multi-agent path finding (MAPF) with communication requires agents to reach individual goals without collisions under partial observability. Learnable policies trained on expert data provide an effective approach to this problem. However, when several coordinated joint actions are vali

中文介绍 面向含通信的去中心化多智能体路径规划(MAPF),在部分可观测下需无碰撞到达目标。当多个协调的联合动作都有效时,去中心化决策易冲突。本文提出通过迭代意图去噪进行联合动作细化(Decentralized Master-Mind),以专家数据训练策略。

Fewer Tokens, Better Action: GPT-6 Astra Robot Agents with 14% Higher Success Rate but 65% Fewer Tokens

👍 34

Vision language model (VLM) agents can control robots through visual feedback and action primitives, but repeated model invocations and redundant observations incur substantial token overhead. We introduce PyRUA-Lean, an interactive code-execution framework that couples feedback-driven primitive com

中文介绍 VLM 智能体控制机器人时常因重复调用与冗余观测产生大量 token 开销。本文提出 PyRUA-Lean 交互式代码执行框架,结合反馈驱动的基元组合,减少模型调用与冗余观察,用更少的 token 实现更高任务成功率。

Architect-Ant: Editable Automatic Furnishing of Architectural Floor Plans

👍 27

Furnished floor plans support real-estate visualization, interior design, and architectural workflows, yet automatic furnishing remains challenged by limited real-world data and the need to satisfy interacting geometric and functional constraints. We ask whether professional furnishing knowledge can

中文介绍 面向建筑平面图自动家具布置(Architect-Ant)。自动布置受限于真实数据稀缺,且需满足几何与功能约束。本文探索能否用专业布置知识生成可编辑的自动布置结果,服务房产可视化、室内设计等建筑工作流。

SILSA: Sliding-Window Slice Latents for Topology-Preserving High-Resolution 3D Generation

👍 27

High-resolution 3D generation increasingly relies on voxel latents and multi-stage pipelines that first predict active structure and then synthesize local geometry. While effective, this design fragments continuous surfaces into many local tokens, inflates generation cost, and often weakens topologi

中文介绍 高分辨率 3D 生成常用体素潜变量与多阶段流水线,但会把连续表面碎片化为本地 token、抬高成本并削弱拓扑。SILSA 提出滑窗切片潜变量(Sliding-Window Slice Latents),在保持拓扑的前提下进行高分辨率 3D 生成,降低片段化与生成开销。

Decoding Looped Transformers Better for (Almost) Free

👍 25

Looped Transformers achieve parameter efficiency by repeatedly executing a shared block across recurrent loops. Each loop yields an intermediate representation decodable for the same next token, yet standard decoding discards earlier states. Because earlier loops embody less computation, recurrence

中文介绍 循环 Transformer 通过反复执行共享块实现参数高效,但标准解码只取最后一次状态,丢弃更早的中间表示。由于早期循环计算更少却对应同一 next token,本文提出改进解码方法,几乎免费地利用各循环中间状态提升输出质量。

CorrGRPO: Correlation-Normalized GRPO for Multi-Reward Learning

👍 24

Group Relative Policy Optimization (GRPO) is widely used to train reasoning language models, where it computes advantages by centering and normalizing rewards across rollouts of the same prompt. For multiple rewards, GRPO sums the reward components and normalizes the total reward by its within-group

中文介绍 GRPO 在多奖励场景下通常先对各奖励求和再按组归一化,掩盖了各奖励间的差异。CorrGRPO 提出相关性归一化的 GRPO,在组内按相关性处理多个奖励分量,使多奖励学习更稳定,改善推理语言模型的训练效果。

Multimodal Flow: Unified Flow Modeling of Language and Vision in Embedding Spaces

👍 23

We present Multimodal Flow, a fully continuous generative model of language and vision. Most unified multimodal models either model both language and quantized images as discrete tokens or combine discrete language prediction with continuous image generation. The former introduces a visual quantizat

中文介绍 提出 Multimodal Flow,一个在嵌入空间对语言与视觉做统一连续建模的生成模型。多数多模态模型将语言与量化图像都作离散 token,或在离散语言预测与连续图像生成间拼合。本文消除视觉量化损失,实现全连续的多模态生成。

Ego2Act: Evaluating Goal-Directed Manipulation in Egocentric Video Generation

👍 21

Video generation models are increasingly being explored as world simulators for embodied planning and learning. To do so effectively, these models must not only generate visually appealing frames, but also predict how environments dynamically evolve when executing goal-directed actions. While evalua

中文介绍 视频生成模型作为具身规划的世界模拟器时,除视觉美观外还须预测执行目标导向动作后环境的动态演化。Ego2Act 基准专评第一人称视频生成中的目标导向操作能力,检验其是否真正模拟环境随动作的动态变化。

Argo-Bench: Evaluating Data Agents on Enterprise-Scale Workflows

👍 21

Real-world enterprise data science and analytics workflows require reasoning across dozens of tables, performing statistical analyses, and acting on the results. Established text-to-SQL benchmarks evaluate query generation alone, and audits have found their answer keys frequently wrong. Because real

中文介绍 面向企业级数据科学与分析工作流,提出 Argo-Bench。真实工作流需跨数十张表推理、统计分析与基于结果采取行动,而现有 text-to-SQL 基准仅评查询生成且答案键常出错。该基准评估数据智能体在真实大规模工作流中的综合能力。

Omni-Embed-Mini: Binding Modalities Without Forgetting via Dense Distillation

👍 19

Extending a text embedding model to new modalities typically degrades text retrieval quality, and existing omni-modal embedders compensate with multi-billion parameters. We present Omni-Embed-Mini, a 0.9B-parameter model that maps text, speech, audio, images, video, and visually-rich documents into

中文介绍 扩展文本嵌入模型到新模态常损害文本检索质量,现有多模态嵌入器依赖数十亿参数。Omni-Embed-Mini 以仅 0.9B 参数将文本、语音、音频、图像、视频及富视觉文档映射到统一空间,借助稠密蒸馏在绑定模态的同时不遗忘文本能力。

PhysVista: Benchmarking Physical Intelligence in VLMs via a Perception-Reasoning-Assessment Loop

👍 17

Vision-Language Models (VLMs) have shown strong multimodal reasoning capabilities, yet whether they truly capture the physical consistency underlying real-world dynamics remains unclear. Existing benchmark paradigms often suffer from fragmented evaluation, focusing on isolated cognitive stages while

中文介绍 检验 VLM 是否真正捕捉现实动态的物理一致性。现有基准评价零散、聚焦单一认知阶段。PhysVista 构建感知-推理-评估闭环基准,评估视觉语言模型在物理智能方面的整体能力,而非孤立考察某一环节。

Better Supervision Is Nearby: Neighborhood On-Policy Self-Distillation

👍 15

On-policy self-distillation (OPSD) trains mathematical reasoning models using a privileged teacher that sees a reference solution and supervises student-sampled prefixes. Standard OPSD uses one fixed parameter setting at every state, but nearby settings may offer additional supervision. We find that

中文介绍 策略自蒸馏(OPSD)用看到参考答案的特权教师监督学生采样的前缀,但每个状态只用固定参数设置。本文发现邻近参数设置可提供额外监督,提出邻域策略自蒸馏(Neighborhood OPSD),用更丰富监督提升数学推理模型效果。

Smaller Models, Better Rejects: Preference Distillation Scaling

👍 15

Preference distillation typically treats a teacher response as preferred and the student's own response as rejected. This assumes that self-generated failures are the most informative negatives and that rejects must come from a model at least as large as the student, making generation costly at scal

中文介绍 偏好蒸馏通常把教师回复当偏好、学生自身回复当拒绝,假设自生成失败最具信息量且拒绝须来自至少与学生同大的模型,导致规模化时生成昂贵。本文探索偏好蒸馏的规模化,用更小的模型产生高质量拒绝样本,降低成本。

The Rise of the Automation Engineer

@jakeserval · 1.1K 粉丝 · 406.0K 阅 · 505 赞 · 1.1K 转

There is a lot of talk about the jobs AI takes away and very little about the ones it creates. While pessimism might be easier (and certainly gets more clicks), I’m more optimistic. From where I sit,

中文介绍 博主认为市场过度强调 AI 抢工作,而忽略其创造的新岗位。他结合自身观察,提出「自动化工程师」是 AI 时代兴起的新角色,核心能力是把重复流程、数据管道、Agent 工作流串联成可落地系统,强调 AI 在创造岗位与重塑工作形态上的潜力。

here's how to make viral videos with Seedance 2.5

@0xROAS · 54.6K 粉丝 · 118.6K 阅 · 502 赞 · 56 转

Seedance 2.5 is one of the craziest "nano banana moments" of AI VIDEO in 2026. It's the model behind every viral video that you are seeing right now... and it's pretty simple to generate those video,

中文介绍 0xROAS 分享用 Seedance 2.5 制作病毒式 AI 视频的实操方法,称其为 2026 年视频生成领域的「nano banana 时刻」。文中拆解生成爆款短视频的工作流,强调该模型操作简单、出片质量高,是当下社媒流传的 AI 视频背后主力引擎。

Own your AI, don’t let it own you

@0xSigil · 56.0K 粉丝 · 79.1K 阅 · 509 赞 · 72 转

6 months ago, the #1 AI was Claude Opus 4.6. Our model, Underdog 27B beats it while running on your Mac or iPhone. Most AI will run on your device, not in data centers. Intelligence that recently

中文介绍 0xSigil 发布端侧模型 Underdog 27B,可在 Mac 或 iPhone 本地运行,宣称性能超过半年前的 Claude Opus 4.6。核心主张是「拥有自己的 AI,别被 AI 拥有」,认为未来多数 AI 将运行在用户设备上而非云端,强调端侧推理的自主性与隐私优势。

Designing MCP Gateway Uber's MCP Management Platform

@UberEng · 66.9K 粉丝 · 54.0K 阅 · 529 赞 · 59 转

Introduction Uber’s rapid adoption of AI agents has fundamentally changed how teams interact with code, data, and operational systems. Early ad-hoc integrations with the MCP (Model Context Protocol)

中文介绍 Uber 工程团队介绍其 MCP 管理平台的设计思路。随着团队大规模采用 AI Agent,代码、数据与运维系统的早期临时集成方式已难维护。文中详述 MCP(Model Context Protocol)网关架构,解决多 Agent 环境下统一连接、权限管控与协议治理问题。

Evals: how to know whether an AI system actually works

@sermakarevich · 549 粉丝 · 33.0K 阅 · 520 赞 · 64 转

An article for everyone who ships, buys, or signs off on software built on large language models (LLMs): engineers, product managers, and the CEO. Written in plain language, from the big picture down

中文介绍 sermakarevich 撰写面向工程师、产品经理及 CEO 的 AI 系统评估(Evals)长文,用平实语言从宏观到细节讲解如何判断一个基于 LLM 的系统是否真正可用。内容涵盖评估框架设计、指标选择、测试流程及常见陷阱,适合已交付或正在采购 AI 产品的团队参考。

How to build a content machine with Opus 5.5 (Full Course)

@VibeMarketer_ · 37.8K 粉丝 · 31.4K 阅 · 504 赞 · 59 转

i'm going to show you how to build a content machine with Opus 5.5 that turns your ideas, research and expertise into articles, posts and newsletters in your own voice. from finding your next topic to

中文介绍 VibeMarketer_ 推出使用 Opus 5.5 搭建内容流水线的完整教程。从选题、研究到产出文章、社媒帖文和 newsletter,全流程用 Opus 5.5 生成并保持个人写作风格。课程覆盖提示词设计、风格迁移、批量生产等环节,目标是让个人创作者高效产出规模化内容。

Reinforcement Learning from scratch #1

@goyal__pramod · 11.4K 粉丝 · 24.7K 阅 · 504 赞 · 59 转

Now I am aware you must have been looking forward to the second part of my CUDA blog (if you haven't read it yet, check it out!), but hey! A man can have varied interests. And I believe if you are

中文介绍 goyal__pramod 发布从零手写强化学习的系列教程第一期。此前他写过 CUDA 博客,本次转向 RL 主题,认为两者都是深度学习基础设施的关键方向。教程从基础概念入手,用代码逐步实现核心算法,适合想深入理解 RL 原理而非仅调用框架的开发者。

OpenAI Withheld GPT-6.1 Astra. It Exposed AI’s Permission Problem.

@ActionModelAI · 60.2K 粉丝 · 6.8K 阅 · 501 赞 · 439 转

OpenAI has chosen not to release GPT-6.1 Astra after the model failed to meet its safety standards. To be clear, GPT-6.1 Astra is not GPT-6.1 Sol, the model OpenAI released at DevDay. GPT-6.1 Sol has

中文介绍 ActionModelAI 讨论 OpenAI 决定不发布 GPT-6.1 Astra 的决策,称该模型因未通过安全标准被搁置,与 DevDay 上发布的 GPT-6.1 Sol 是不同版本。文章借此分析 AI 的「权限问题」,探讨前沿模型在能力与安全性之间的取舍及其对行业发布的启示。

The Agent Said It Was Done. The Database Disagreed.

中文介绍 微软在Hugging Face发布博文,讨论名为ThinkingBox的智能体系统:智能体报告任务已完成,但数据库核验显示并未实际写入,揭示AI代理在数据库操作可靠性上的缺口。

[AINews] not much happened today

a quiet day.

中文介绍 Latent Space的AINews栏目发布当日简讯,称当天AI领域无重大进展,内容以一句「nothing much happened」总结,属于低信息量的日报格式。

A model guide for the GPT-6 family

Learn how startups can choose GPT-6 models, tune reasoning effort, improve prompts and skills, coordinate tools, and prepare workflows for production.

中文介绍 OpenAI发布GPT-6系列模型选型指南,面向初创公司讲解如何在推理强度、提示调优、工具协同与生产化工作流之间平衡选择,覆盖多个GPT-6模型变体。

Redefining enterprise intelligence with autonomous AI

Enterprise AI is no longer a future ambition. It is in full operational flight. Model capabilities are advancing faster than most organizations can absorb, while the cost of performance continues to fall. Globally, AI investment is set to reach $2.5 trillion in 2026, up 44% from the previous year. F

中文介绍 MIT科技评论指出企业AI已全面落地,模型能力增速超过组织消化速度,成本持续下降;预计2026年全球AI投资达2.5万亿美元,同比增长约44%。

Inside-Out AI: Rebuilding Airbnb Behind the Scenes and Across the Guest Experience

After leading Meta’s Llama models, Ahmad Al-Dahle is now transforming Airbnb with AI — from how its teams develop products to how it serves guests.

中文介绍 前Meta Llama团队负责人Ahmad Al-Dahle加入Airbnb,推动该公司从内部产品开发流程到面向客人的服务体验的全栈AI改造。

Don’t be fooled—LLMs don’t reason

On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board in what looked like a gift to its human opponent. Move 37 in game two of the five-game match looked so absurd that some commentators thought it was a…

中文介绍 MIT科技评论以2016年首尔一场围棋比赛中AlphaGo第37手的争议性落子为例,质疑大语言模型是否真正具备推理能力,认为「LLM并不真正推理」。

Academia is for Ambition — Alex Zhang, MIT

We catch up with RLM first author Alex Zhang, MIT PhD, on Jev, PhD masxing, and the future of harnesses.

中文介绍 Latent Space专访MIT博士生、RLM论文一作Alex Zhang,聊及Jev现象、博士生涯加速以及未来「harness」工具的走向。

Chatham scales its capital markets expertise with OpenAI

Chatham Financial uses Codex and GPT-5.6 to build technology and redesign workflows, cutting trade validation from 30 minutes to under 4.

中文介绍 Chatham Financial使用OpenAI的Codex与GPT-5.6重建交易验证流程,将单笔交易校验时间从约30分钟压缩至4分钟以内。

The eternal complement

Advanced AI may matter most for the routine work behind breakthrough ideas. Explore why execution could shape the next economy and the pace of progress.

中文介绍 OpenAI发文讨论先进AI在突破性想法背后的执行环节中的价值,认为执行力可能决定下一个经济周期与技术进步节奏。

How Albertsons Companies is reimagining retail from the inside out

Albertsons Cos. is using ChatGPT Enterprise and the OpenAI API to help teams work faster and make grocery shopping easier for millions of customers.

中文介绍 Albertsons公司借助ChatGPT Enterprise与OpenAI API提升内部协作效率,并改善数百万消费者的杂货购物体验。

[AINews] Gemini 4 Argon: GDM’s answer to Astra/Fable, with 1M output

... but you can’t try it yet unless you are “government users and trusted cyber defenders in the Fairwind Program”

中文介绍 GDM推出Gemini 4 Argon模型,对标Astra/Fable系列,最大输出长度达100万token,但仅限Fairwind计划内政府用户及受信任的网络安全防御者试用。

市场总览

美股四大科技板块整体维持多头排列:SPY 收于 769.64(距 52 周高仅 -1.25%),QQQ 749.58(-0.66%),NVDA、MSFT、AAPL 均位于 20/50/200 日均线之上,RSI 多集中在 54–65 中性偏强区间,MACD 出现分歧(QQQ/NVDA/MSFT 金叉上行,SPY/AAPL 死叉回落),短线动能有走弱迹象但趋势仍为牛市。加密市场总市值 2.89 万亿美元(24h -1.88%),BTC 主导率 59.1%、ETH 11.4%,恐慌贪婪指数 65 处于「贪婪」区间;BTC 85262.61、ETH 2702.30、SOL 121.42 均呈 SMA20>SMA50>SMA200 多头排列,RSI 在 62–66 健康偏强但未超买。中概股集体走熊:BABA/PDD/0700.HK/JD 全线空头排列,RSI 30–41 逼近超卖,MACD 全部死叉。商品外汇方面,黄金 4162.3 空头排列、RSI 34.1;WTI 原油 91.11 短线承压但均线仍多头;美元/人民币 6.70 接近 52 周低、空头排列。宏观层面,10Y 美债收益率 5.28% RSI 74.6 超买,美元指数 101.92 RSI 71.7 同样超买且接近 52 周高,VIX 15.31 回落 6.59% 显示风险偏好修复。整体呈现「美股韧性 + 加密偏热 + 中概承压 + 美债美元超买」的四分格局。

今日关注

TSLA 特斯拉 (TSLA)
偏上行

现价 370.59,单日大涨 4.65%,已上穿 SMA20 364.87 与 SMA50 347.58,但仍受压于 SMA200 393.18;RSI14 反弹至 55.0 脱离弱势区,MACD 1.8335 虽较信号 3.332 偏低仍处死叉状态,但接近 52 周高 -25.71% 的位置显示中短期修复动能,空头排列有松动迹象。

^TNX 10Y 美债收益率
偏上行

10Y 收益率升至 5.28%,单日 +0.76%、5 日 +1.79%,站上 SMA20 5.05、SMA50 4.83、SMA200 4.45 形成标准多头排列;RSI14 高达 74.6 已进入超买区,MACD 0.1343 略高于信号 0.1163 保持红柱,距 52 周高仅 -1.22%,上行趋势明确但短期超买压力上升。

DX-Y.NYB 美元指数 DXY
偏上行

现价 101.92,单日 -0.17% 但 5 日仍 +0.94%,价格远高于 SMA20 100.36、SMA50 100.00、SMA200 99.27,多头排列完好;RSI14 高达 71.7 已超买,MACD 0.5794 持续高于信号 0.3881 红柱扩张,距 52 周高 -0.28% 触及前高区域,趋势偏强但短期存在回踩需求。

0700.HK 腾讯控股 (0700.HK)
偏下行

现价 421.20,单日 -2.27%、5 日 -3.92%,价格已跌破 SMA20 433.7、SMA50 447.51、SMA200 498.88 三重均线,形成空头排列;RSI14 仅 40.6 弱势但未及超卖,MACD -3.9952 持续低于信号 -3.9536 死叉延续,距 52 周低仅 +2.48% 逼近历史低位,下行压力明显。

GC=F 黄金期货
偏下行

现价 4162.3,单日 -0.95%、5 日 -3.68%,价格低于 SMA20 4334.33、SMA50 4366.27、SMA200 4553.37 完整空头排列;RSI14 34.1 接近超卖但未反弹,MACD -63.4969 大幅低于信号 -38.6231 绿柱扩张,距 52 周高 -25.49% 深度回撤,空头动能仍然占优。

全部资产

^VIX

VIX 恐慌指数

$15.31 -6.59%
5 日
+2.96%
距 52w 高
-56.6%
RSI(14)
47.6
趋势
空头
SMA 20 / 50 / 200
15.92 / 15.78 / 18.06
MACD / 信号
0.005 / -0.027
MACD 金叉 (3 天前)空头排列

^TNX

10Y 美债收益率 (%)

$5.28 +0.76%
5 日
+1.79%
距 52w 高
-1.2%
RSI(14)
74.6
趋势
多头
SMA 20 / 50 / 200
5.05 / 4.83 / 4.45
MACD / 信号
0.134 / 0.116
RSI 超买接近 52 周高多头排列

DX-Y.NYB

美元指数 DXY

$101.92 -0.17%
5 日
+0.94%
距 52w 高
-0.3%
RSI(14)
71.7
趋势
多头
SMA 20 / 50 / 200
100.36 / 100.00 / 99.27
MACD / 信号
0.579 / 0.388
RSI 超买接近 52 周高多头排列

SPY

S&P 500 ETF

$769.64 +0.74%
5 日
-0.22%
距 52w 高
-1.2%
RSI(14)
54.5
趋势
多头
SMA 20 / 50 / 200
764.83 / 763.70 / 720.47
MACD / 信号
0.836 / 0.893
MACD 死叉 (2 天前)接近 52 周高多头排列

QQQ

Nasdaq 100 ETF

$749.58 +1.02%
5 日
+0.68%
距 52w 高
-0.7%
RSI(14)
65.4
趋势
多头
SMA 20 / 50 / 200
727.78 / 716.82 / 668.60
MACD / 信号
8.205 / 6.488
接近 52 周高多头排列

AAPL

Apple

$333.69 +1.02%
5 日
-2.16%
距 52w 高
-3.4%
RSI(14)
54.7
趋势
多头
SMA 20 / 50 / 200
331.89 / 322.42 / 289.16
MACD / 信号
3.807 / 4.820
MACD 死叉 (3 天前)多头排列

MSFT

Microsoft

$517.53 +0.92%
5 日
+0.26%
距 52w 高
-6.5%
RSI(14)
62.6
趋势
多头
SMA 20 / 50 / 200
501.67 / 487.72 / 432.83
MACD / 信号
7.804 / 7.445
MACD 金叉 (1 天前)多头排列

NVDA

Nvidia

$233.95 +1.34%
5 日
+3.95%
距 52w 高
-1.7%
RSI(14)
62.9
趋势
多头
SMA 20 / 50 / 200
223.79 / 218.28 / 200.71
MACD / 信号
3.489 / 2.661
接近 52 周高多头排列

GOOGL

Alphabet

$343.50 +1.56%
5 日
-0.12%
距 52w 高
-15.9%
RSI(14)
50.0
趋势
中性
SMA 20 / 50 / 200
342.62 / 344.37 / 338.92
MACD / 信号
-0.546 / -0.397
MACD 死叉 (1 天前)

TSLA

Tesla

$370.59 +4.65%
5 日
-0.41%
距 52w 高
-25.7%
RSI(14)
55.0
趋势
中性
SMA 20 / 50 / 200
364.87 / 347.58 / 393.18
MACD / 信号
1.834 / 3.332
MACD 死叉 (4 天前)

META

Meta

$728.08 +0.30%
5 日
-3.14%
距 52w 高
-6.6%
RSI(14)
61.5
趋势
中性
SMA 20 / 50 / 200
695.92 / 625.04 / 628.38
MACD / 信号
34.313 / 35.083
MACD 死叉 (今天)
加密恐慌贪婪
65
贪婪
加密总市值
$2.89 T
-1.88% / 24h
BTC 主导率
59.1%
ETH 11.4%
24h 成交量
$44.9 B
活跃币 21,833

BTC-USD

Bitcoin

$85,262.61 +0.59%
5 日
+1.96%
距 52w 高
-32.4%
RSI(14)
65.3
趋势
多头
SMA 20 / 50 / 200
82,796.82 / 79,003.42 / 71,495.53
MACD / 信号
1,917.057 / 2,058.280
多头排列

ETH-USD

Ethereum

$2,702.30 +0.56%
5 日
+0.96%
距 52w 高
-43.2%
RSI(14)
61.9
趋势
多头
SMA 20 / 50 / 200
2,646.73 / 2,500.52 / 2,118.30
MACD / 信号
64.076 / 75.284
多头排列

SOL-USD

Solana

$121.42 +1.48%
5 日
+1.99%
距 52w 高
-48.8%
RSI(14)
66.0
趋势
多头
SMA 20 / 50 / 200
115.03 / 104.69 / 85.95
MACD / 信号
4.960 / 5.337
MACD 死叉 (3 天前)多头排列

BABA

阿里巴巴 (BABA)

$105.85 -1.49%
5 日
-3.54%
距 52w 高
-44.2%
RSI(14)
35.6
趋势
空头
SMA 20 / 50 / 200
110.07 / 116.87 / 130.93
MACD / 信号
-2.389 / -2.184
MACD 死叉 (2 天前)空头排列

PDD

拼多多 (PDD)

$75.38 -1.46%
5 日
-2.82%
距 52w 高
-45.9%
RSI(14)
33.4
趋势
空头
SMA 20 / 50 / 200
78.47 / 83.77 / 94.50
MACD / 信号
-1.874 / -1.841
MACD 死叉 (今天)空头排列

JD

京东 (JD)

$25.78 -2.24%
5 日
-2.75%
距 52w 高
-28.7%
RSI(14)
30.4
趋势
中性
SMA 20 / 50 / 200
26.93 / 29.01 / 28.89
MACD / 信号
-0.661 / -0.660
MACD 死叉 (今天)

0700.HK

腾讯控股 (0700.HK)

HK$421.20 -2.27%
5 日
-3.92%
距 52w 高
-38.3%
RSI(14)
40.6
趋势
空头
SMA 20 / 50 / 200
433.70 / 447.51 / 498.88
MACD / 信号
-3.995 / -3.954
MACD 死叉 (今天)接近 52 周低空头排列

GC=F

黄金期货

$4,162.30 -0.95%
5 日
-3.68%
距 52w 高
-25.5%
RSI(14)
34.1
趋势
空头
SMA 20 / 50 / 200
4,334.33 / 4,366.27 / 4,553.37
MACD / 信号
-63.497 / -38.623
空头排列

CL=F

WTI 原油期货

$91.11 -1.90%
5 日
-1.41%
距 52w 高
-23.7%
RSI(14)
47.4
趋势
多头
SMA 20 / 50 / 200
96.04 / 88.58 / 82.38
MACD / 信号
0.552 / 1.821
多头排列

USDCNY=X

美元 / 人民币

¥6.70 +0.00%
5 日
-0.12%
距 52w 高
-6.8%
RSI(14)
39.8
趋势
空头
SMA 20 / 50 / 200
6.71 / 6.73 / 6.82
MACD / 信号
-0.006 / -0.007
接近 52 周低空头排列
风险提示

本报告仅基于公开行情数据的技术指标读数进行客观描述,不构成投资建议。技术指标反映历史价格结构,过去走势不代表未来表现。本报告仅供技术指标解读参考,投资者应自行评估风险并结合基本面信息独立决策。

Friedrich Merz begins unannounced Kyiv visit as Russia continues strikes on Ukraine’s capital – Europe live

German chancellor tours city with president Zelenskyy after Moscow strikes city’s North Bridge this morning amid his arrival Merz travelling to Kyiv this weekend is an important display of support, especially as the bombardments intensify. As Pjotr Sauer also noted in his report: Western diplomats s

中文摘要 德国总理默茨未预先通知突访基辅,在俄罗斯军方于其抵达当日对基辅北桥发动袭击的背景下,与泽连斯基共同视察该市。此行被视为对乌克兰的重要支持信号。

The story of a book in Gaza: A treasured possession given new lease of life

With libraries, schools and homes destroyed in Gaza, surviving books have become treasured possessions.

中文摘要 加沙地带大量图书馆、学校和民宅遭毁后,幸存的书籍成为珍贵物品。报道讲述了一本图书在废墟中被重新赋予生命的故事。

默茨访问乌克兰许诺增加援助 俄军继续轰炸基辅大桥

德国总理默茨未事先宣布突然对乌克兰进行了访问,承诺将进一步向基辅提供新的军事援助。不过,德国民调显示,支持扩大对乌军援的德国人口正逐年下降。

中文摘要 德国总理默茨突然访问乌克兰,承诺向基辅提供更多军事援助。但民调显示,德国国内支持扩大对乌军援的比例正逐年下降。

Who Were the Pilots on the FlyDubai Flight?

The co-pilot’s embrace of extreme Islamist views prompted Omani officials to scrutinize him, according to two people briefed on the investigation.

中文摘要 两名知情人士透露,迪拜飞航空难中副机长的极端伊斯兰思想导致阿曼当局对其进行审查。

What to Know About Fighting Between Ethiopia and Rebels in Tigray

The conflict is raising fears of a regional war in the Horn of Africa, drawing in foreign powers including Eritrea, Egypt, the United Arab Emirates and Saudi Arabia.

中文摘要 埃塞俄比亚与提格雷反叛武装冲突升级,引发东非之角地区战争担忧。厄立特里亚、埃及、阿联酋和沙特等域外大国正介入其中。

Photos: Community unites in Senegal’s Dakar to clean dilapidated school

Volunteers and students prepare safer, cleaner classrooms in advance of the academic year in Senegal's capital.

中文摘要 塞内加尔首都达喀尔社区居民与志愿者在学年开始前齐心协力,清理一所破旧的校舍,以提供更安全、更干净的教室。

Who Owns the Moon?

A treaty from the 1960s declared that nobody can own outer space. Half a century later, billionaires and superpowers see fortunes to be made nonetheless.

中文摘要 上世纪60年代的《外空条约》规定无人可拥有太空。半个世纪后,亿万富翁与大国仍视月球为巨大商机,月球主权争议重新升温。

U.K. Base Would Be Ambitious Target for ‘Gig Economy’ Warfare

The U.S. war in Iran unleashed small-scale attacks by local criminals or disenfranchised young men hired anonymously to target Jewish and American sites in Europe and North America, experts say.

中文摘要 专家认为,美国对伊朗的战争催生了「零工经济式战争」,雇佣当地罪犯或边缘青年匿名袭击欧美犹太和美国目标,英国基地可能成为雄心勃勃的攻击对象。

Brazil Election: What to Know as Lula and Bolsonaro Face Off in 2026 Presidential Race

Luiz Inácio Lula da Silva, the leftist incumbent, and Flávio Bolsonaro, the right-wing challenger, face off in a race that could define the future of Latin America.

中文摘要 巴西2026年总统大选拉开帷幕,左翼现任总统卢拉与右翼挑战者弗拉维奥·博索纳罗正面交锋。这场高度极化的选举可能决定拉丁美洲的未来走向。

Nobel Prizes 2026: What to Know

Six awards will be announced this week in science, literature, economics and peace work.

中文摘要 本周将公布2026年诺贝尔奖六项奖项,涵盖科学、文学、经济学和和平奖等领域。

Russians Snap Up Chinese E.V.s as Ukrainian Attacks Make Fuel Scarce

As Ukraine increasingly takes the war to Russia, the effects may be most visible on the roads.

中文摘要 随着乌克兰将战火引入俄罗斯本土,炼油厂遭袭导致俄境内燃油短缺,俄罗斯消费者转而抢购中国电动车。

The Search for Gas in Moscow

Our reporter Ivan Nechepurenko looks at gas stations in Moscow to see how severe the shortages have become as Ukrainian attacks target Russian refineries.

中文摘要 《纽约时报》记者内切普留连科走访莫斯科加油站,实地探访乌克兰对俄炼油厂打击后油料短缺的严重程度。

Superpowers Race to Put Nuclear Reactors on the Moon

The United States wants a reactor on the moon by 2030. A Russian-Chinese alliance is working on one for 2036. Some leading scientists say the danger is great.

中文摘要 美国计划在2030年前将核反应堆送上月球,俄中联盟的目标为2036年。一些顶尖科学家警告,此举风险巨大。

Brazil goes to polls in decisive election for future of Latin America

Lula and Flávio Bolsonaro face off in battle between leftwing incumbent and far-right son of jailed ex-president Brazilians go to the polls on Sunday in the first round of a highly polarised and high-stakes presidential election that could change not only the direction of Latin America’s largest cou

中文摘要 巴西2026年总统大选首轮投票举行,左翼现任总统卢拉对阵在押前总统博索纳罗之子弗拉维奥·博索纳罗,这场高度极化的选举可能改变拉丁美洲走向。

Why the IPO market is booming and busting

Flotations have stalled and returns from new listings have been weak

中文摘要 FT分析全球IPO市场:新股发行明显降温,近期多笔上市项目被暂停,二级市场新上市股票回报率疲弱,投资者需求不足与估值担忧成为主要制约因素。

Australian authorities probe Flydubai attacker’s links to the country

Recriminations among various governments have escalated after the attempted hijacking of a flight from Dubai to Tel Aviv last week

中文摘要 澳大利亚当局就上周迪拜飞往特拉维夫航班遭劫持未遂事件展开调查,重点排查袭击者与澳洲的关联。多国政府互相指责,外交摩擦持续升级。

Sahel juntas launch TV channel with Russian backing

Military governments in Mali, Burkina Faso and Niger bet broadcaster will bolster popularity as they fight Islamist insurgencies

中文摘要 马里、布基纳法索和尼日尔三国军政府获得俄罗斯支持,联合创办电视台,意图借媒体宣传提升在打击伊斯兰极端叛乱中的民意支持度。

OPEC+ Set to Keep Quotas Steady in November, Delegates Say

Major OPEC+ nations have an agreement in principle to keep oil production quotas unchanged next month, as conflict in the Middle East continues to shutter swaths of the group’s output.

中文摘要 彭博援引OPEC+代表消息,主要产油国已原则上同意11月维持石油产量配额不变。中东冲突持续压制该地区输出,市场关注减产与供应风险。

Ethiopia Retakes Tigray Airport as Northern Rebels Retreat

Ethiopian forces seized control of the airport and other parts of Mekelle, the capital of the northern Tigray region, as leaders of the Tigray People’s Liberation Front retreated from the city, a major turning point in the renewed conflict.

中文摘要 埃塞俄比亚军队重新控制提格雷地区首府麦凯勒的机场及市区多处区域,提格雷人民解放阵线领导人撤离,被视为该轮冲突的关键转折点。

Wall Street’s IPO fervour cools on tepid demand and valuation worries

Several listings have been paused in recent weeks as delay in Anthropic’s public debut sends a chill through markets

中文摘要 FT报道华尔街IPO热潮降温:近几周多笔上市被推迟,Anthropic上市计划延迟对市场情绪造成冲击,需求低迷与估值忧虑成为主要阻力。

Andy Burnham and the art of corporate pitch-rolling

The UK prime minister has many grand plans, but most won’t kick in until 2030 or even later

中文摘要 FT评论英国首相安迪·伯纳姆的企业投资拉拢策略:其政府有大量宏伟产业与基建计划,但多数项目要到2030年甚至更晚才启动见效。

Masayoshi Son’s AI ambitions outgrow SoftBank’s balance sheet

DigitalBridge CEO Marc Ganzi says his data centre investment group will be SoftBank’s ‘third-party infrastructure arm’ after $4bn takeover

中文摘要 FT报道孙正义的AI投资规模已超出软银资产负债表承载能力。DigitalBridge CEO Marc Ganzi称其数据中心投资集团将成为软银40亿美元收购后的第三方基础设施臂。

How Renault went from driveshafts to drones

French carmaker turns mass-manufacturing expertise to new frontier of warfare

中文摘要 FT分析雷诺转型:这家法国车企正将整车大规模制造能力延伸到军用无人机领域,从传统汽车供应链切入国防与战争科技新赛道。

Google set to defend £1.2bn UK lawsuit over ‘excessive’ app download charges

British claim over Google Play Store is latest effort to challenge market power of Big Tech groups

中文摘要 谷歌将于英国法庭应诉一项12亿英镑诉讼,指控其在Google Play商店对应用下载收取过高费用,此举被视为挑战科技巨头市场权力的最新努力。

China closes hundreds of banks to bolster financial system

More than 670 lenders, a record, shut down last year as Fitch says smaller players remain sector’s weakest part

中文摘要 FT报道中国去年关闭超过670家银行,创历史纪录,以强化金融体系。惠誉称小型金融机构仍是该行业最薄弱环节,风险尚未完全出清。

Japanese and Korean shipbuilders deploy robots to take on China

US allies invest in ‘smart shipyards’ as Beijing’s dominance of industry alarms Washington

中文摘要 FT报道日韩船厂投资「智能造船厂」并部署机器人,以应对中国在造船业的领先地位。美国盟友加大投入,北京在该行业的优势令华盛顿警觉。