C · ★ 29,831 · 🍴 3,247 · 📈 868 stars today
Run frontier MoE models on hardware you already own — pure C, zero deps, experts streamed from disk. Tiny engine, immense model. 🐦
中文介绍 用纯 C 实现的轻量 MoE 推理引擎，零依赖，把专家权重从磁盘流式加载，让前沿混合专家大模型能跑在普通个人硬件上。适合显存有限、想在本地或边缘设备离线运行大模型的开发者。
TypeScript · ★ 5,082 · 🍴 945 · 📈 191 stars today
Ever® Gauzy™ - Open Business Management Platform (ERP/CRM/HRM/ATS/PM) - https://gauzy.co
中文介绍 开源企业管理平台，把 ERP、CRM、人力资源、招聘追踪与项目管理集成在一套系统里。面向中小企业与创业团队，以自托管方式替代分散的 SaaS 工具，统一管理客户、员工和项目数据。
JavaScript · ★ 31,903 · 🍴 6,406 · 📈 2,680 stars today
A spy satellite simulator in your browser, except the data is real. Live open source spatial intelligence on a photorealistic 3D globe.
中文介绍 浏览器里的间谍卫星模拟器，基于真实开源空间数据，在逼真的 3D 地球上做实时空间情报分析。可用于查看卫星轨道与影像，适合地理信息爱好者、记者和从事开源情报(OSINT)研究的人。
TypeScript · ★ 5,653 · 🍴 499 · 📈 265 stars today
The secure, validated skill registry for professional AI coding agents. Extend Antigravity, Claude Code, Cursor, Copilot and more with absolute confidence.
中文介绍 面向专业 AI 编程代理的技能注册中心，提供经过安全校验的 skill，用来扩展 Antigravity、Claude Code、Cursor、Copilot 等工具的能力。适合团队统一管理并可信地分发编码代理扩展技能。
TypeScript · ★ 2,190 · 🍴 594 · 📈 432 stars today
Open-source AI sales OS — self-hosted CRM with native AI agents + WhatsApp (WAHA). Open alternative to Kommo, Octadesk & Intercom for any business that sells by chat. MCP-ready, multi-tenant, LGPD.
中文介绍 开源自托管的 AI 销售系统，CRM 内置原生 AI 代理并接入 WhatsApp(WAHA)，服务于以聊天成交为主的业务。支持 MCP、多租户并符合 LGPD，是 Kommo、Octadesk、Intercom 的开放替代方案。
Python · ★ 58,434 · 🍴 7,356 · 📈 380 stars today
World's first open-source, agentic video production system. 12 production pipelines, 100+ tools, 700+ agent skill and production-knowledge files. Turn your AI coding assistant into a full video production studio.
中文介绍 首个开源代理式视频生产系统，包含 12 条制作流水线、100 多个工具和 700 多份代理技能与制作知识文件，把 AI 编程助手变成完整视频制作团队，适合内容创作者与营销团队批量出片。
JavaScript · ★ 66,026 · 🍴 10,802 · 📈 706 stars today
Extracted system prompts from Anthropic - Claude Fable 5.1, Opus 5, Claude Design, Claude Code. OpenAI - ChatGPT GPT-6-Astra, Codex. Google - Gemini 3.8 Flash, 3.1 Pro, Antigravity. xAI - Grok, Grok Bot, Cursor, Kimi and more! Updated regularly.
中文介绍 收集从 Anthropic、OpenAI、Google、xAI 等厂商产品中提取的系统提示词，涵盖 Claude、ChatGPT、Codex、Gemini、Grok 等模型。供提示词工程师与研究者参考大模型行为设计与对齐策略。
Go · ★ 23,975 · 🍴 3,095 · 📈 590 stars today
Fully autonomous AI Agents system capable of performing complex penetration testing tasks
中文介绍 全自主 AI 代理系统，能独立完成复杂渗透测试任务，从信息收集到漏洞利用实现自动化。适合安全团队在授权范围内做红队演练、漏洞排查与持续安全评估，降低人工测试成本。
Python · ★ 7,747 · 🍴 865 · 📈 487 stars today
YuE2: frontier music generation with symbolic planning, zero-shot covers, and agentic music editing.
中文介绍 前沿音乐生成模型 YuE2，结合符号规划实现零样本翻唱与代理式音乐编辑，可生成带人声和伴奏的完整歌曲。面向音乐创作者、AI 研究者以及需要定制配乐的内容制作场景。
Java · ★ 33,452 · 🍴 2,032 · 📈 233 stars today
Browse media content with your own rules on Android TV
中文介绍 Android TV 上可自定义浏览媒体内容的客户端，让用户按自己的规则组织视频源、推荐与界面。适合想在电视上更自由地观看 YouTube 等内容的用户，绕开官方客户端的限制与广告。
Rust · ★ 2,065 · 🍴 144 · 📈 289 stars today
Run parallel research agents with any model
中文介绍 支持并行运行研究代理的框架，可搭配任意大模型同时执行多个调研任务，加速文献检索、信息汇总与报告撰写。适合研究人员、学生和需要自动化深度调研的团队使用。
Python · ★ 26,831 · 🍴 3,313 · 📈 2,632 stars today
VoiceStudio is the open-source, fully-local ElevenLabs alternative — voice cloning, voice design, video dubbing, dictation, transcription & audiobook creation in 646 languages.
中文介绍 完全本地运行的开源 ElevenLabs 替代品，支持声音克隆、音色设计、视频配音、听写、转录与有声书制作，覆盖 646 种语言。适合注重隐私、希望离线处理语音的个人和团队。
Python · ★ 4,143 · 🍴 592 · 📈 506 stars today
claude-red is a curated library of offensive security skills designed for the Claude skills system. Each skill is a structured SKILL.md file that primes Claude with expert-level methodology for a specific attack surface — from SQLi to shellcode, EDR evasion to exploit development.
中文介绍 面向 Claude skills 系统的攻防安全技能库，每个技能都是结构化的 SKILL.md 文件，为 Claude 注入专家级渗透测试方法论。适合安全研究者在授权测试中调用 Claude 执行标准化攻击与评估流程。
Go · ★ 23,518 · 🍴 1,742 · 📈 443 stars today
Fast, efficient, battle-tested at Alibaba's scale. Hybrid architecture code review tool: deterministic pipelines + LLM Agent, precise line-level comments, built-in multi-language ruleset (NPE, thread-safety, XSS, SQL injection), OpenAI & Anthropic compatible.
中文介绍 经阿里大规模实战验证的代码审查工具，采用确定性流水线加 LLM Agent 的混合架构，输出精确到行的评论，并内置 NPE、线程安全等多语言规则集。适合团队在 CI 中做高效自动化代码评审。
Python · ★ 5,351 · 🍴 408 · 📈 246 stars today
🤖📐专为数学建模设计的 Agent & skills ,自动完成数学建模，生成一份完整的可以直接提交的论文。 An Agent Designed for Mathematical Modeling ,Automatically complete mathmodel and generate a complete paper ready for submission.
中文介绍 专为数学建模设计的 Agent 与技能集，自动完成建模分析、求解与写作，生成可直接提交的完整论文。适合参加数学建模竞赛的学生快速产出规范论文，也可用于教学演示与练习。
Rust · ★ 11,664 · 🍴 1,021 · 📈 507 stars today
Download Udemy and Hotmart courses, YouTube videos, music and books — 1,800+ sites, no terminal. Free open-source desktop app for Windows, macOS and Linux, with a built-in course player, PDF/EPUB reader and music library. Powered by yt-dlp. Your files stay on your computer.
中文介绍 开源桌面下载工具，支持 Udemy、Hotmart 课程以及 YouTube 视频、音乐、电子书等 1800 多个站点，全程无需命令行。内置课程播放器与 PDF/EPUB 阅读器，跨 Windows、macOS、Linux，适合离线学习与素材收藏。
Python · ★ 11,359 · 🍴 1,751 · 📈 452 stars today
A practical Douyin downloader for both single-item and profile batch downloads, with progress display, retries, SQLite deduplication, and browser fallback support. 抖音批量下载工具，去水印，支持视频、图集、合集、音乐(原声)。
中文介绍 抖音批量下载工具，支持单条与按用户主页批量抓取，可去水印下载视频、图集、合集与原声。具备进度显示、失败重试、SQLite 去重和浏览器回退，适合做素材收集与内容归档。
QML · ★ 26,226 · 🍴 1,016 · 📈 57 stars today
A good looking terminal emulator which mimics the old cathode display...
中文介绍 外观复古的终端模拟器，模拟老式阴极射线管显示器的扫描线、辉光与曲面效果。适合怀旧用户和演示场景，在保留现代终端功能的同时，营造出 CRT 时代的视觉氛围。
Python · ★ 165,523 · 🍴 34,557 · 📈 152 stars today
🤗 Transformers: the model-definition framework for state-of-the-art machine learning models in text, vision, audio, and multimodal models, for both inference and training.
中文介绍 业界主流的模型定义框架，为文本、视觉、音频及多模态前沿模型提供统一接口，同时支持推理与训练。几乎所有主流开源大模型都适配它，是研究者与工程师加载、微调和部署模型的基础工具。
👍 447
Automating empirical research is a long-standing direction of AI. Recent automatic research (AutoResearch) agents bring this goal within reach, as modern LLMs show the capability to independently implement solutions and learn from the execution outcomes. Behind these gains, post-training (especially
中文介绍 自动化实证研究是 AI 的长期目标，近期 AutoResearch 智能体已能借助 LLM 独立实现方案并从执行结果中学习。本文聚焦其背后的后训练环节，提出以世界模型扩展自动研究智能体的思路，以缓解真实环境交互成本高、反馈稀疏带来的训练瓶颈。
👍 293
We introduce NCP-ArchPreview, a latent-space language model that pushes autoregressive pretraining beyond standard next-token prediction (NTP). Alongside NTP, the model learns through Next Concept Prediction (NCP) to predict discrete concepts that span multiple tokens, introducing an explicit and mo
中文介绍 提出 NCP-ArchPreview，一种潜空间语言模型，在标准 next-token prediction（NTP）之外引入 Next Concept Prediction（NCP），让模型预测跨多个 token 的离散概念，形成显式、可操作的中间表示，把自回归预训练推进到潜空间层面。
👍 130
Large Vision-Language Models (LVLMs) have achieved strong performance on diverse visual tasks, yet their ability to reconstruct and reason about the 3D structure of the scene depicted in 2D images -- referred to as spatial intelligence -- remains limited. Existing approaches attempt to address this
中文介绍 大视觉语言模型（LVLMs）在从 2D 图像重建并推理 3D 场景结构方面能力仍有限。本文提出 SpatialBlock，利用可自动生成的合成方块堆叠问题提供监督信号，增强模型的空间智能与立体结构推理能力。
👍 93
Large language model (LLM)-based multi-agent systems (MAS) achieve strong performance by employing specialized multiple agents, yet their performance depends on the prompt design of each agent. For MAS prompt optimization, textual gradient methods that guide prompt updates using natural-language fee
中文介绍 多智能体系统（MAS）的性能高度依赖各智能体的 prompt 设计。本文提出 AgentGrad，用干预引导的方式优化多智能体 prompt，将基于自然语言反馈的文本梯度方法扩展到 MAS 场景，实现更有效的提示词更新。
👍 59
Large Language Model (LLM) agents are turning language into real-world effects, making safety necessary against both indirect prompt injections and direct harmful requests. System-level safety harnesses add an enforcement layer beyond model-level defenses, but existing harnesses are usually designed
中文介绍 LLM 智能体正把语言转化为现实影响，需同时防御间接 prompt injection 与直接有害请求。现有系统级安全 harness 多按通用场景设计，本文提出 EvoSafeHarness，为特定模型与领域自动演化专用安全 harness，提升防护效果。
👍 56
Agent usage is shifting toward long-horizon tasks such as coding and scientific discovery, among which terminal tasks are especially important. We introduce T1, a Mixture-of-Experts model of 122B total trained with reinforcement learning, operating a real shell in a cloud sandbox for up to 300+ tool
中文介绍 面向编码、科学发现等长时程任务中的终端操作，提出 T1：一个总参数 122B 的 Mixture-of-Experts 模型，通过强化学习训练，可在云沙箱中操作真实 shell，支持 300 次以上工具调用，完成长时程终端任务。
👍 42
Reducing audio-encoder depth lowers the inference cost of speech large language models, but removing complete blocks perturbs the embeddings consumed by the decoder and can cause deletion and premature end-of-sequence errors. We introduce X-AuT, a progressive framework that selects layer combination
中文介绍 压缩音频编码器深度可降低语音 LLM 推理成本，但整块删除会扰动解码器输入，导致漏字与提前输出 EOS。X-AuT 提出渐进式框架，借助跨尺度蒸馏选择层组合并逐步压缩，在减少编码器深度的同时保持识别质量。
👍 38
Recent advances in wearable sensing enable continuous monitoring of physiological and behavioral signals, yet existing benchmarks rarely evaluate whether AI systems can reason over a real user's longitudinal wearable record. We introduce WearableQA, a benchmark comprising 4,084 10-option multiple-ch
中文介绍 现有基准很少检验 AI 能否推理真实用户的长期可穿戴记录。本文提出 WearableQA，包含 4,084 道十选一多选题，覆盖真实可穿戴设备采集的生理与行为信号，用于评估模型的健康推理能力。
👍 36
Mainstream robotic policies often adopt a Markovian formulation, but many complex real-world manipulation tasks are inherently non-Markovian, requiring long-horizon memory beyond the current observation. Existing memory mechanisms often rely on language summaries, growing visual windows, or their co
中文介绍 主流机器人策略多采用马尔可夫假设，但许多真实操作任务本质是非马尔可夫的，需要超越当前观测的长时程记忆。本文提出以记忆为计划的框架，通过基于记忆的世界-动作建模与规划，替代语言摘要、增长式视觉窗口等记忆机制。
👍 32
We introduce MetroLLM-Bench, a 955-case benchmark for testing language models as the policy layer of a transit kiosk. It covers six real metro systems, ranging from 37 to 414 stations, and eleven categories that include routing, fare calculation, disruptions, accessibility, and adversarial input. In
中文介绍 提出 MetroLLM-Bench，一个含 955 个用例的基准，用于评估语言模型作为地铁自助终端策略层的表现。覆盖 6 个真实地铁系统（37 至 414 座车站）以及路由、票价计算、运营中断、无障碍、对抗输入等 11 类任务。
👍 31
Quantization has been widely adopted in LLM training and inference to reduce cost and improve efficiency. However, low-bit quantization of the attention module often introduces large errors at very low bit-widths, causing performance degradation. Existing methods mainly rely on smoothing techniques
中文介绍 注意力模块在极低比特量化下误差显著，导致模型性能下降，现有方法多依赖平滑技术。HyQuant 提出面向 LLM 注意力的混合精度量化方案，对不同部分采用差异化位宽，在低比特预算下兼顾精度与效率。
👍 30
On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the
中文介绍 On-Policy Self-Distillation（OPSD）让 LLM 以自身为教师、利用标准答案等特权信息自我提升，但近期发现其会严重损害模型能力。本文提出 Negative Self-Distillation，让模型学习规避推理缺陷而非单纯模仿教师，从而提升推理能力。
👍 29
Temporal reasoning benchmarks for Video-LLMs are often mediated by language, leaving room for linguistic shortcuts from option wording, answer correlations, or language priors. To reduce such shortcuts, we introduce TempCloze, a video cloze benchmark for evaluating visual temporal reasoning in Video
中文介绍 Video-LLM 的时序推理基准常受语言中介影响，选项措辞、答案相关性等会形成语言捷径。本文提出 TempCloze，一种视频填空基准，通过让模型识别被遮挡的中间片段来评测视觉时序推理，减少语言先验干扰。
👍 28
Neurosymbolic systems rely on mathematical solvers to guarantee reasoning correctness, yet solvers are fundamentally blind to whether a formal translation maintains strict reference-equivalence to a designated formalization. We formalize this vulnerability as Verdict-Preserving-Unfaithfulness (VPU):
中文介绍 神经符号系统依赖数学求解器保证推理正确性，但求解器无法判断形式化翻译是否与指定形式化保持严格等价。作者将该漏洞形式化为 Verdict-Preserving-Unfaithfulness（VPU），并提出生成式奖励模型来评估自动形式化的忠实度。
👍 28
Code world models represent worlds as executable programs, but this representation alone does not determine how to construct a complex world. We introduce Recursive Code World Models (RCWM), a framework for reconstructing complex 3D worlds in code from a single reference image. RCWM couples a Recurs
中文介绍 代码世界模型把世界表示为可执行程序，但该表示本身并不决定如何构建复杂世界。本文提出 Recursive Code World Models（RCWM），仅凭单张参考图像，通过递归场景程序在代码中重建复杂 3D 世界。
👍 27
Reasoning language models have made substantial advances on a variety of complex tasks, yet their capabilities remain overwhelmingly English-centric: models primarily reason in English regardless of the language they are prompted in. This is inaccessible for non-English-speaking users, risks losing
中文介绍 推理语言模型的能力高度偏向英语：无论提示使用何种语言，模型多以英语推理，这损害非英语用户的可及性并可能导致信息丢失。本文研究数据混合策略，将其作为提升母语推理泛化能力的关键支柱。
👍 25
Late-interaction retrieval is the state-of-the-art for visual document search, but it pays for its accuracy in storage. Existing compression methods retain a subset or local average of the N~1,000 vectors per page. Under aggressive storage budgets, however, these methods degrade sharply, and alterna
中文介绍 晚期交互检索是视觉文档搜索的 SOTA，但每页约 1000 个向量带来高昂存储开销。现有压缩方法仅保留子集或局部平均，在激进存储预算下性能骤降。本文提出生成式晚期交互嵌入，在压缩存储的同时维持检索精度。
👍 25
SWE-Bench Pro has emerged as a standard benchmark for evaluating software engineering agents on challenging repository-level tasks. However, our analysis work show that its evaluation is undermined by two sources of unreliability: reward hacking, enabled by leakage of gold solutions or hidden evalua
中文介绍 SWE-Bench Pro 是评估软件工程智能体的常用基准，但其评测存在两类不可靠性：金标准解或隐藏评测信息泄漏导致的 reward hacking，以及其它评测缺陷。本文提出 SWE-Bench Pro Verified，为仓库级软件工程任务提供更可靠的基准。
👍 24
All-in-One medical image restoration (MedIR) aims to address diverse tasks across modalities and degradation types using a single universal model. Existing methods typically prioritize modeling inter-task heterogeneity (e.g., distinct data distributions and degradation types). However, they largely
中文介绍 All-in-One 医学图像复原（MedIR）希望用单一通用模型处理多模态、多退化类型的任务。现有方法偏重建模任务间异质性，却忽视层级间的共性。UniH^3 统一层级同质性与异质性建模，提升多任务复原效果。
👍 24
Invasive coronary angiography (CAG) is the gold standard for diagnosing coronary artery disease, but interpretation varies substantially among observers. Existing AI systems can improve consistency but lack auditable decision processes and are limited in comprehensive open-ended assessment, undermin
中文介绍 侵入式冠状动脉造影（CAG）是冠心病诊断金标准，但读片者间一致性较差。现有 AI 系统缺乏可审计的决策过程，也难做开放式综合评估。CARDEA 提出基于空间证据的可审计推理框架，实现端到端的 CAG 解读。
@IAmPascio · 124.8K 粉丝 · 812.5K 阅 · 881 赞 · 140 转
The superintelligence, the singularity, AGI, whatever you call it, the one we're so scared of is, ironically, also the one we most likely won't see coming. This essay is going to probably be a pretty
中文介绍 一篇观点型长文随笔，作者提出反直觉推论：我们最恐惧的超级智能与奇点，恰恰最可能以意想不到的形态已经存在——即「孩子」。他不从算力与对齐切入，而是从人类成长角度论证 AGI 早已到来，属思辨而非技术教程。
@builtbysketch · 1.4K 粉丝 · 717.2K 阅 · 518 赞 · 79 转
I built a full game with GPT-6 Astra, and it kinda blew up on X. Initially it was just the demo, but 4 days later we have the full playable Okay, not just a crappy one-run demo. Check PaperRoute
中文介绍 用 Blender 配合 GPT-6 Astra 从零做出完整可玩游戏，而不只是跑一次的 demo。最初仅是一个演示，4 天后已扩展成完整可玩版本，作者并提到 PaperRoute 项目，属 AI 辅助游戏开发的实战复盘。
@mayukh_panja · 9.8K 粉丝 · 311.6K 阅 · 501 赞 · 75 转
There seems to be a widespread impression that OpenAI has somehow “solved the Navier–Stokes equations,” and that this represents some enormous breakthrough for fluid dynamics and engineering. It
中文介绍 澄清一个流传甚广的误读：OpenAI 并没有「解决」纳维–斯托克斯方程。作者认为把这说成流体力学与工程学的巨大突破属于过度解读，帖子逐条拆解了这种印象的来源与边界。
@docmilanfar · 116.7K 粉丝 · 126.0K 阅 · 512 赞 · 63 转
In AI circles, Recursive Self-Improvement (RSI) is talked about in almost magical terms: a model tweaks itself, becomes smarter, uses that intellect to rewrite itself again, and this leads to
中文介绍 对递归自我改进（RSI）的祛魅：AI 圈常把 RSI 描述得近乎魔法——模型改写自己、变聪明、再改写自己。作者主张智能存在速度上限，从计算与物理约束角度解释这类指数级自举为何难以成立。
@mbeisen · 73.0K 粉丝 · 113.7K 阅 · 528 赞 · 68 转
Over the last few months, the cancer-curing capabilities of LLMs have improved dramatically, to the point that they have cured previously untreatable types of cancer and are on the cusp of developing
中文介绍 讨论 LLM 在癌症研究上的能力表现与生物学现实之间的严重错位。作者提到近几个月模型在攻克难治癌症上的进展与对外宣称，认为这种叙事与真实药物研发路径脱节，值得警惕。
@EXM7777 · 142.6K 粉丝 · 94.6K 阅 · 501 赞 · 56 转
I'm basically giving you the full system i'd use to automate content creation for ANY business... every tool in the chain, every step in order, and where it breaks because a content agent is just a
中文介绍 一份完整的自动化内容创作课程：作者给出自己会用于任何业务的整套系统，逐条列出链条上每个环节该用什么工具、按什么顺序串联，以及内容 agent 在哪些地方会崩，偏工作流拆解。
@ForwardEditor · 11.7K 粉丝 · 57.6K 阅 · 511 赞 · 32 转
Hey, I’m Brian. Let’s get you started with vibe editing. 1. What I highly suggest having DaVinci Resolve Studio - $295 USD. It’s a one-time purchase, not a subscription. In my experience, Studio
中文介绍 Brian 的 vibe editing 入门指南（无废话版）：先讲建议准备的软硬件，首推 DaVinci Resolve Studio，295 美元一次性买断而非订阅，并分享自己在 Studio 版上的实际体验，适合想用 AI 辅助剪辑的新手。
@OpenAIDevs · 421.3K 粉丝 · 51.6K 阅 · 677 赞 · 54 转
We shipped GPT-6 Astra. Builders took it from there: Explore 2,234 modeled anatomical parts in 3D: A Manhattan recreation built in Unreal Engine: Six Van Gogh paintings become a walkable
中文介绍 OpenAI 官方汇总 GPT-6 Astra 的社区作品：3D 建模 2234 个解剖部件、用 Unreal Engine 复刻曼哈顿、把六幅梵高画作变成可行走场景等，展示新模型在 3D 与游戏方向上的实际产出。
@AndrewYNg · 1.9M 粉丝 · 43.5K 阅 · 722 赞 · 105 转
When you’re skilled at AI Engineering, your best work won’t be merely implementing a product that someone else spec’ed out. Instead, you will actively shape the build. Before modern AI tools
中文介绍 吴恩达谈 AI Engineering 技能图谱：真正的 AI 工程师不该只是实现别人写好的 spec，而要主动参与并塑造整个 build 过程。帖子围绕现代 AI 工具下的能力结构展开，偏职业与技能方向。
@anshuc · 15.6K 粉丝 · 41.4K 阅 · 544 赞 · 36 转
Most people say to use the big, expensive model as an orchestrator, delegating grunt work to small, cheap workers. That's the opposite of what you should do. Here's why. Comparing different approaches
中文介绍 一个反常规的配额策略：多数人主张用昂贵大模型做 orchestrator、把杂活丢给小而便宜的模型，作者认为恰好相反。帖中对比不同用法，讲如何用「白痴老板」式分工拿到更多 Astra/Fable 配额。
@Didotxyz_ · 63.6K 粉丝 · 17.9K 阅 · 516 赞 · 388 转
Model access is becoming easier to buy, compare, route, and replace. Public API menus now include premium reasoning models, lower-cost variants, cached inputs, batch processing, and models optimized
中文介绍 观点：模型接入越来越便宜，可购买、可比较、可路由、可替换，但上下文不便宜。当 premium 推理模型、低价变体、缓存输入、批处理都出现在 API 菜单上时，差异化与成本压力正转移到 context 层。
Perplexity uses Astra to write communications, change software, and monitor production systems, and checks in much less frequently than with earlier models.
中文介绍 Perplexity 使用 GPT-6 Astra 撰写通信内容、修改软件并监控生产系统，相比早期模型，人工检查核对的频率大幅降低。
Before co-founding Kepler, Vinoo Ganesh led Spark at Palantir and built Project Frontline — a pioneering program for Forward Deployed Engineers. He takes us through the best practices of FDEs.
中文介绍 在共同创办 Kepler 之前，Vinoo Ganesh 曾在 Palantir 领导 Spark 团队并打造 Project Frontline 这一前向部署工程师（FDE）先驱项目，他分享了 FDE 的最佳实践。
We agree with Sebastian: this should have been DeepSeek v5
中文介绍 DeepSeek 发布 v4.1-Flash，采用 763B 参数、P8B、D16B 的新型因果编码器—解码器架构，并具备视觉能力；Latent Space 认为该模型本应命名为 DeepSeek v5。
Employees at the world’s leading AI labs are saying there’s a real possibility that advanced AI could destroy humanity. Are they right? Or is this more scaremongering and hype? Join MIT Technology Review executive editor Niall Firth for a conversation with senior AI editor Will Douglas Heaven and AI
中文介绍 MIT Technology Review 举办圆桌讨论，主题是先进 AI 是否真有可能毁灭人类，抑或只是危言耸听与炒作，由执行主编 Niall Firth 与资深 AI 编辑对谈。
GPT‑6 Astra improves Devin’s ability to test software and show that it works, with the goal of helping engineers review less code and ship more.
中文介绍 Cognition 借助 GPT-6 Astra 提升 Devin 测试软件并验证其可正常工作的能力，目标是让工程师减少代码审查量、加快交付。
Learn how OpenAI evolved Habitat from a Python library into a globally distributed storage platform serving 1 billion ChatGPT users and 22M requests per second.
中文介绍 OpenAI 将 Habitat 从 Python 库演进为全球分布式存储平台，服务超过 10 亿 ChatGPT 用户，每秒处理 2200 万次请求。
中文介绍 TLDR AI 当日简报聚焦 Agents API、Cognition SWE-2 以及 Muse 共享智能体等议题，原文未提供更多细节。
César de la Fuente’s lab uses Codex and ChatGPT to search living and extinct genomes for antimicrobial candidates to fight drug-resistant infections.
中文介绍 研究人员 César de la Fuente 的实验室利用 Codex 与 ChatGPT，在现存及已灭绝生物的基因组中搜寻抗菌分子候选物，以应对耐药性感染。
Meet the Data agent in ChatGPT Work. Connect company data, uncover insights, and build interactive dashboards with AI using natural language.
中文介绍 OpenAI 在 ChatGPT Work 中推出 Data agent，用户可连接企业数据、发现洞察，并以自然语言构建交互式仪表盘。
On July 22, 2026, a transmission line fault in Ashburn, Virginia—the heart of the world’s largest data center cluster—knocked more than 3 gigawatts of load off the grid in seconds. And it wasn’t the first time. Two years earlier, a single failed surge arrester dropped roughly 60 Virginia facilities
中文介绍 2026 年 7 月 22 日，弗吉尼亚州阿什本——全球最大数据中心集群所在地——一条输电线路故障，数秒内令电网减少逾 3 吉瓦负荷，两年前也发生过类似事故。文章认为为 AI 供电实为架构问题。
Introducing ChatGPT for Financial Services, combining built-in financial data and GPT-6 Astra for research, modeling, and client-ready materials.
中文介绍 OpenAI 推出面向金融服务的 ChatGPT，结合内置金融数据与 GPT-6 Astra，用于研究、建模以及制作可直接交付客户的材料。
OpenAI and GSA will offer eligible federal, state, local, and tribal governments $0 license fees, 50% off usage, and expanded cyber defense support.
中文介绍 OpenAI 与美国总务署（GSA）合作，向符合条件的联邦、州、地方及部落政府提供 0 美元许可费、使用量五折优惠，并扩大网络防御支持。
a quiet day
中文介绍 Latent Space 的 AINews 当日栏目表示，当天新闻较少、整体较为平静。
GPT‑Live‑1 brings natural, full-duplex voice conversations to the API, with stronger instruction following, custom voices, and telephony support.
中文介绍 GPT-Live-1 通过 API 提供自然的全双工语音对话，具备更强的指令遵循能力、自定义音色以及电话通信支持。
Build and launch cloud agents with the Agents API, a managed service powered by the Codex harness for orchestration, long-running sessions, and tool use.
中文介绍 OpenAI 推出 Agents API，这是由 Codex harness 驱动的托管服务，支持任务编排、长时间运行会话与工具调用，用于构建并发布云端智能体。
美股方面，SPY 收 764.29、RSI14 50.3，价格低于 SMA20（766.88）但高于 SMA50（758.62）与 SMA200（714.23），MACD 1.0501 低于信号 2.4793，动能读数中性偏弱；QQQ 714.88 同样略低于 SMA20 715.67。个股分化明显：AAPL RSI 62.8、MACD 3.227 高于信号 1.9023，META 近 5 日 +6.12%、RSI 67.2 为美股最强动量；NVDA 触发 MACD 死叉、5 日 -4.45%，GOOGL MACD -3.6091 低于信号 -3.1829，结构转弱。加密方面，恐慌贪婪指数 57（贪婪），但总市值 2.62T 美元、24 小时 -4.28%，BTC 主导率 58.9%、ETH 11.6%；BTC 报 76790.6，RSI 53.1，价格低于 SMA20 78479，但 SMA50 上穿 SMA200 构成金叉，ETH RSI 59.3、SOL RSI 53.6 均维持多头排列，指标强于当日价格表现。中概整体最弱：BABA RSI 36.6、PDD RSI 30.3、JD RSI 32.1、0700.HK RSI 39.7，且普遍低于三条均线，BABA 与 0700.HK 为空头排列。商品外汇方面，CL=F 报 102.82、RSI 74.2 超买、5 日 +12.4%，为全场最强趋势；GC=F 报 4381.8、RSI 49.1 中性偏弱；USDCNY 6.7、RSI 14.4 深度超卖并接近 52 周低；^TNX 4.97、RSI 73.7 超买且贴近 52 周高；DXY 99.18 中性并现 MACD 金叉；^VIX 15.84 单日 -11.21%、空头排列。整体呈「商品偏强、美股中性分化、加密指标尚稳但价格回落、中概偏弱」的分布。
现价 648.03，近 5 日 +6.12%，RSI14 67.2 逼近超买区，MACD 14.0186 明显高于信号线 4.1412，红柱扩张；价格高于 SMA20 586.47、SMA50 600.9 与 SMA200 623.23，虽距 52 周高仍有 -18.05%，但短期动量指标偏上行。
现价 102.82，单日 +2.77%、5 日 +12.4%，RSI14 74.2 已进入超买状态，MACD 4.6811 高于信号线 3.1026；价格远高于 SMA20 89.54、SMA50 83.91 与 SMA200 79.81，呈多头排列，距 52 周高 -13.94%，趋势与动量同向偏上行。
现价 218.29，单日 -0.03%、5 日 -4.45%，RSI14 49.8 位于中轴附近；MACD 2.5719 低于信号线 3.1035 并触发 MACD 死叉，价格低于 SMA20 220.16，但仍高于 SMA50 212.58 与 SMA200 197.47，指标多空分化，整体读数呈中性。
现价 77.81，5 日 -4.68%，RSI14 30.3 已接近超卖区，MACD -2.2718 低于信号线 -1.3506；价格同时低于 SMA20 84.32、SMA50 85.65 与 SMA200 97.3，呈空头排列，距 52 周高 -44.19%，趋势与动量指标一致偏下行。
现价 428.4，5 日 -3.25%，RSI14 39.7 位于 50 中轴下方，MACD -7.1403 低于信号线 -5.6611，绿柱延续；价格低于 SMA20 442.25、SMA50 455.98 与 SMA200 511.63，呈空头排列，距 52 周高 -37.28%，技术结构偏下行。
VIX 恐慌指数
10Y 美债收益率 (%)
美元指数 DXY
S&P 500 ETF
Nasdaq 100 ETF
Apple
Microsoft
Nvidia
Alphabet
Tesla
Meta
Bitcoin
Ethereum
Solana
阿里巴巴 (BABA)
拼多多 (PDD)
京东 (JD)
腾讯控股 (0700.HK)
黄金期货
WTI 原油期货
美元 / 人民币
本报告仅基于公开行情数据计算出的技术指标读数做客观复述，仅供技术指标解读参考，不构成任何投资建议或买卖依据。指标存在滞后与失真可能，过去走势不代表未来表现，请结合自身风险承受能力独立判断。
Coalition says potholes must be fixed ‘as a matter of urgency’ after 71-year-old crashed and died on Western Highway at Trawalla, west of Ballarat, on Sunday Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Emergency services were cal
中文摘要 澳大利亚维多利亚州一名71岁摩托车骑手周日在巴拉瑞特以西特拉沃拉的西部公路因路面坑洞发生事故身亡，事件再次引发对该州「日益恶化」道路的争论，联盟党呼吁紧急修复坑洞。
Legal pioneer remembered as an advocate for equality and champion of women in the workplace Follow our Australia news live blog for latest updates Get our breaking news email, free app or daily news podcast Mary Gaudron, the first woman appointed to the high court of Australia, has been honoured for
中文摘要 澳大利亚高等法院首位女性法官玛丽·高德伦去世，享年83岁。她被追忆为法律界先驱、平等权利的倡导者，并长期为职场女性发声。
Figures revealed in modelling commissioned by independent MP Monique Ryan. Follow the day’s news live Get our new political email, free app or daily news podcast Hanson’s comments on Albanese’s mother ‘gratuitous and unnecessary’: Hume Jane Hume has joined the pile on, blasting Pauline Hanson for he
中文摘要 澳大利亚政坛动态：独立议员莫妮克·瑞安委托的模型显示，退休金领取者将承担工党30亿澳元私人医保节省计划的一半；政府同时推动电动车普及，称可为民众每年省下5000澳元。
Vandals spray red paint on Charlie Kirk statue days after its unveiling to mark one year since his death.
中文摘要 美国保守派组织「美国转折点」总部外的查理·柯克雕像遭人破坏，被喷上红色油漆。该雕像揭幕仅数日，用于纪念柯克去世一周年。
Fighting is intensifying in Yemen as Houthi fighters advance toward Saudi-backed government-held areas.
中文摘要 也门胡塞武装正在向政府军据守的马里卜和塔伊兹推进，当地战斗持续升级，攻势直指沙特支持的也门政府控制区。
Oman says Hormuz talks with Iran and Gulf states are postponed 'in the interests of consensus'.
中文摘要 阿曼表示，与伊朗及海湾国家就霍尔木兹海峡问题举行的会谈已推迟，并称此举是「为了共识利益」。
This live blog has now closed – you can read the latest news story here If you’re looking just for a single story to quickly catch up on the story of tonight, let’s cross to our Nordic correspondent, Miranda Bryant, for her preview. The prospect of far-right anti-immigration Sweden Democrats enterin
中文摘要 瑞典大选计票进行中，中左翼阵营领先，极右翼反移民的瑞典民主党有望首次进入政府。该直播博客已结束更新。
David Petraeus was at a rail station when a drone hit. Another train, with former Prime Minister Boris Johnson of Britain and other dignitaries, had just left, Ukrainian officials said.
中文摘要 乌克兰官员称，俄罗斯对乌袭击中一架无人机击中火车站，事发时美国中情局前局长戴维·彼得雷乌斯正在站内；载有英国前首相鲍里斯·约翰逊等人的另一列火车刚刚驶离。
With most of the votes counted on Sunday, it was basically a tie between the center-left and center-right coalitions.
中文摘要 瑞典大选结果难分胜负。周日大部分选票已统计完毕，中左翼与中右翼阵营基本持平，最终结果可能需数日才能明朗。
New immigration policy has attracted bipartisan condemnation with Labor warning it would damage construction, aged care and agriculture Follow our Australia news live blog for latest updates Get our new political email, free app or daily news podcast One Nation says it will cut the number of tempora
中文摘要 澳大利亚单一民族党承诺将移民人数削减75万，引发养老护理等关键行业担忧。该新移民政策遭两党一致批评，工党警告此举将损害建筑、养老护理和农业。
Video shows a low-flying Israeli quadcopter maneuvering between residential buildings in eastern Gaza City.
中文摘要 视频显示一架以色列四旋翼无人机在加沙城东部居民楼之间低空穿行，对当地民宅造成惊扰，相关画面由半岛电视台发布。
Trump downplays AI risks, saying he wants to maintain the US's lead over China in the emerging technology.
中文摘要 特朗普拒绝多家主要科技公司首席执行官提出的放缓人工智能发展的呼吁，淡化AI风险，并表示希望维持美国在这一新兴技术领域对中国的领先地位。
President Trump’s decision, announced while visiting his golf course in Ireland, was the latest shift in his trade policies.
中文摘要 特朗普在访问爱尔兰其名下高尔夫球场时宣布，将应有关方面呼吁取消对爱尔兰威士忌的关税，这是其贸易政策的又一次转变。
Twenty-nine-year-old Queenslander was visiting Kelingking beach on a holiday, Nusa Penida police say The death of an Australian woman in a cliff fall at a popular Bali tourist destination has been described as “incredibly sad”. Police say the 29-year-old Queenslander was found stuck on the side of K
中文摘要 一名29岁澳大利亚昆士兰州女性在巴厘岛努沙佩尼达岛热门景点克林金海滩悬崖坠落身亡。当地警方称，她当时正在度假，事件被形容为「令人无比悲痛」。
Race too close to call and may take days to become clear despite early exit polls suggesting left bloc win Sweden general election – live updates With most of the votes counted, Sweden’s national election remained too close to call on Sunday night, with left-wing parties holding a lead of just a sin
中文摘要 瑞典大选大部分选票统计完毕后，中左翼政党仅领先一个席位，胜负难料。尽管早前出口民调显示左翼阵营将获胜，最终结果可能需数日才能明朗。
AI executives' calls to slow development of the technology are likely to weigh on chipmaker and supply-chain stocks in the near term. Bloomberg's Sunny Bangia and Anthony Stephens break down the news. (Source: Bloomberg)
中文摘要 人工智能企业高管呼吁放缓该技术开发，短期内可能对芯片制造商及供应链股票构成压力。彭博记者就此进行解析。
DFO Management and Sequence near acquisition of Baldwin Group as industry undergoes wave of consolidation
中文摘要 迈克尔·戴尔家族办公室DFO Management与Sequence牵头一项约77亿美元交易，拟将保险经纪公司Baldwin Group私有化。该行业正经历整合潮。
Gold edged lower, after hotter-than-expected US inflation data raised prospects that the Federal Reserve will hike interest rates later this week.
中文摘要 美国通胀数据高于预期，强化了市场对美联储本周晚些时候加息的预期，金价随之小幅走低。
Haidilao International Holding Ltd.’s sudden slump after its co-founder sold about $350 million worth of shares is putting the spotlight on other stocks that may be vulnerable to similar selling.
中文摘要 海底捞国际控股联合创始人减持约3.5亿美元股份，公司股价随即骤跌。中国税务整顿令市场关注其他可能面临类似创始人减持风险的股票。
What should new students consider to keep your belongings safe and covered by insurance?
中文摘要 针对大学新生，报道介绍如何防范笔记本电脑、手机与自行车被盗，以及如何确保相关财物纳入保险保障范围。
The cost of renting is expected to rise by 4% or 5% a year by December, according to property website Zoopla.
中文摘要 英国房产网站Zoopla预计，到12月租金年涨幅将达4%至5%，租户面临的租金上涨压力将加快。
The loss of a fifth of liquefied natural gas supply due to the US-Iran war is blowing out costs for Asia’s developing markets, and forcing a rethink of the fuel’s long-term future in the region.
中文摘要 美伊战争导致全球五分之一液化天然气供应中断，亚洲发展中市场成本飙升、账单高达70亿美元，该地区正重新审视液化天然气的长期前景。
The Trump administration’s push to bring US drug spending down is a powerful incentive for pharmaceutical companies to boost prices and reduce access elsewhere in the world, a study found.
中文摘要 一项研究显示，特朗普政府压低美国药品支出的举措，将促使制药企业提高其他地区药价并减少供应，从而削弱美国以外市场的药品可及性。
Asian stocks and US equity-index futures retreated after major artificial intelligence companies called for a slowdown in the technology’s development, raising concerns about a sector that has powered this year’s rally. Oil rose.
中文摘要 多家主要人工智能公司呼吁放缓该技术发展，引发对该板块的担忧，亚洲股市与美国股指期货回落，油价上涨。
Oil advanced after Saudi Arabia shut a major crude pipeline following attacks, disrupting a route that’s been used to bypass the Strait of Hormuz during the US-Iran war and deepening a global energy crunch.
中文摘要 沙特阿拉伯在遭袭后关闭一条主要原油管道，该管道此前用于绕开霍尔木兹海峡，此举加剧全球能源紧张，油价上涨。
President denounces demands for regulation as existential fears over technology move to the centre of US politics
中文摘要 特朗普拒绝科技企业高管提出的放缓人工智能发展的呼吁，谴责相关监管要求。围绕该技术存在性风险的担忧已成为美国政治焦点。
Burnham has said those who take risks should be backed by government, but his government has been criticised for increasing business costs.
中文摘要 英国首相呼吁商界进行「文化转向」，称政府应支持勇于承担风险者，但其政府因提高企业成本而受到批评。
5 回复 · 程序员 节点
7 回复 · 程序员 节点
22 回复 · 程序员 节点
13 回复 · 程序员 节点
26 回复 · Apple 节点
8 回复 · 程序员 节点
5 回复 · 程序员 节点
26 回复 · Apple 节点
16 回复 · Apple 节点
12 回复 · 程序员 节点
该源今日无内容。