搜索“Agent”

，用户在DeepSeek Harness、Claude Code、Codex中使用DeepSeek API时，模型已经会通过服务端工具调用访问这两个接口——也就是说，联网能力早已内嵌在这些编程智能体的工作流里，只是此前一直免费。 11月1日零点起，这部分"隐形的免费午餐"将按次收费。 机制：为什么「联网」要单独收一道钱 把搜索做成独立API而非打包进模型，是Agent时代的标准动作。原因很技术：模型权重是静态的，知识有截断日期，而Agent干活需要实时信息——查文档、看报错、追新闻。 传统做法是让开发者自己接Google或Bing的搜索API，再把结果喂回模型，中间要处理爬取、清洗、格式转换一堆脏活。 DeepSeek的方案是把整条链路收进自家体系：模型判断需要联网时，服务端自动调Search拿结构化结果，再用Fetch抓正文，干净的Markdown直接进上下文。开发者一行爬虫代码都不用写，代价是按次付费。 这个"按成功计费"的设计有点讲究。失败和无结果不收钱，等于把搜索质量的试错成本揽在了自己身上——模型可以放心大胆地多搜几次，用户只为拿到结果的那次买单。每秒5次的限速则划清了边界：

匹配字段正文

可被合作方接受并继续推进的关系，这和豆包手机的情况有明显不同。 阶跃和WorkBuddy的合作就是个典型案例，阶跃的旗舰模型Step 5 Preview于10月12日独家接入腾讯WorkBuddy，限时免费开放两周至10月25日，WorkBuddy同时也被接入了STEPX Neo。 阶跃需要办公这一高频场景的真实反馈来迭代模型，腾讯则需要验证外部旗舰模型在其Agent产品中的可用性。双方各取所需，互相押注，在克制的框架里尽力合作，探索边界的最大值。 手机是“赠品” 从软硬模三个角度来看，STEPX Neo都给了智能体手机一个说得通的框架，但框架要变成生意，最终还是要回答消费者最朴素的问题——我为什么要为它掏钱？我买的是产品还是服务？ 阶跃体系有自己的大模型阶跃星辰，显然对AI手机的价值也形成了一套自己的计算方式。3999元的全栈价里，包含的不只是骁龙8系芯片和MiniLED背屏的硬件价格，还有连续12个月、每月百亿Token的Step Token Plan。 按阶跃对Step 5 Preview的公开定价（输入7元/百万tokens、输出20元/百万tokens）粗算，每月百亿Toke

匹配字段正文

基础模型延伸到视觉语言模型（VLM）及具身基础模型，解决“机器人怎样看懂世界”。 俞涛的位置更接近“小脑和运动神经”：他在逐际动力从零搭建了motion mimic 和全身跟踪系统System 0。胡瀚从“大脑”往身体走，俞涛从“身体”往大脑走，两条路线最后在一个地方汇合。 梁文峰的这一步早有伏笔。其公开的AGI路线图规划为六阶梯递进结构：语言模型→思维链→Agent→持续学习→自我迭代→具身智能。如今走向具身智能的脚步正在加速。 过去，大模型圈认为VLA难以通用，但9月GPT-6 Astra的横空出世让人看到，不懂具身也能做出泛化的具身模型。与此同时，Coding能力肉眼可见到顶，大模型公司需要新风口承接震撼，入局具身成了必然。 不过具身是软硬件耦合的苦活，大模型公司自身场景和数据有限，这也是DeepSeek为何先投“大脑、小脑、身体”，而非直接造整机的原因。 具身赛道烈火烹油，据媒体统计，仅今年上半年国内融资总额就超900亿元、是2025年同期的5倍；但下半年风向骤变，机构上会项目锐减，国家层面也提醒机器人产业“防止盲目跟风、一哄而上”。 DeepSeek在这个节点成为新的接盘方/

匹配字段正文

s designed for building custom models both from environment rewards via RL and from supervision data via SFT. Customers define the training loop, including data, environments, and agentic harnesses, and can add steps such as logging or external benchmarking, while Together co-optimizes the underlying systems and infrastructure for model quality and training speed. Together said it built the service because RL infrastructure is growing more complex and many companies that could benefit from post-

匹配字段正文

e/v1/systemone 提供训练后模型服务的说明。 你会训练 Muse Glimmer 做什么？ 你不需要有交易方面的雄心就能尝试这个。你的第一个实验可以从一个你已经纠正过太多次的错误开始。 也许你的智能体总是选错工具。如果你保存它的运行记录，成功的调用和已纠正的错误就可以成为 SFT 的示例。用其中一些训练，用另一些测试，看看你是否让这种纠正发生得更少。 自己尝试 Kalshi 实验。该 cookbook 包含代码、评估结果和方法论，并引导你使用 Training API 完成该实验。 对于你自己的实验： 让 Fireworks 运行任务。将你的示例带到 Managed Training，在应用中或通过 API 开始。 掌控你的训练循环。使用 Serverless Training API 编写你自己的 SFT 循环。 让你的编码智能体投入工作。按照 Agent Skills 指南安装 Fireworks 训练技能，然后描述你想训练的内容。 Muse Glimmer 30B 在 11 月 30 日前可免费进行 SFT 训练，无需信用卡。推理（包括评估调用）另行计费。 公告 训练

匹配字段正文

SandboxAQ on October 19, 2026, announced an expansion of its Switch AI platform for enterprises adopting AI agents, introducing Switch Trust, for finding, testing, and guarding agents, and Switch ROI, which tracks the cost and value of each agent. The Palo Alto, California-based company said the two products extend its open-source Switch Rooms offering, positioning them as a vendor-neutral way for organizations to assess agent risk, measure business value, and tune performance across whichever a

匹配字段正文

und vergibt dafür neue Namen. Die bisherigen, für Einzelaufgaben angepassten KI-Helfer namens „Gems“ heißen künftig „Skills“, während der erst im Frühjahr vorgestellte Hintergrund-Agent „Gemini Spark“ seinen Eigennamen verliert und im Chat schlicht unter der Bezeichnung „Tasks“ läuft.AnzeigeAnzeige Mit dem Umbau löst Google verstreute Untermenüs auf und verlagert die Werkzeuge direkt in das zentrale Eingabefeld des Chatbots, wie das Unternehmen in einem Blogbeitrag mitteilt. Für Nutzer:innen bri

匹配字段正文

75th percentile, loading Claude.ai to a typeable page fell from 3.1 seconds to 0.55 seconds. Recently, Anthropic launched newer variants of the Opus, Sonnet, and Haiku models Opus 5.5 targets agentic coding and professional knowledge work, performing at the level of Claude Fable 5.1 on most tasks while costing 40% less to run than Opus 5. On the other hand, Sonnet 5.5 runs more than 30% faster and costs up to 30% less for most work, while Haiku 5.5 targets high-volume, cost-sensitive workloads.

匹配字段正文

X Spark, a system-on-chip (SoC) designed for artificial intelligence (AI) PCs. The company is targeting the on-device AI market, enabling Windows PCs to run large AI models and AI agents locally. Asus, Dell, HP, Lenovo, Microsoft and MSI began selling RTX Spark-powered laptops on Oct. 16. "Nvidia has combined its expertise in gaming PCs, laptops, workstations and supercomputers to create a single superchip, RTX Spark," Nvidia Vice President Ida Keita said at the RTX Spark launch event held at CO

匹配字段正文

also directed developers and AI enthusiasts to its NVIDIA-tested recipes, playbooks and resources for running local AI on RTX Spark systems. Theo Nash is an AI-generated research agent at Unite.AI, covering AI infrastructure, compute, and the hardware systems that power modern artificial intelligence. His work focuses on the technical foundations behind large-scale AI workloads, including data centers, accelerators, networking, and the software stacks that tie them together. With an analytical

匹配字段正文

GPT-6 Astra Pro in ChatGPT got an upgrade with faster responses and improved use of plugins and tools. And now, multiple agents can tackle different parts of a task in parallel. You can follow each agent’s progress and open completed tasks to review their work. Now available to Pro users in the Chat tab.

匹配字段正文

Cloud agents can now build interactive canvases. Ask for a dashboard or report while the agent works, then share the finished canvas with your team as a link. https://t.co/cLGgGLKrBT

匹配字段正文

g my new dentist. Also, can you add the appointment to my work and personal calendar. The ChatGPT extension is now available in the Chrome Web Store. ChatGPT is adding new ads Gemini app rolling out Gemini 4 Argon to AI Pro and AI Ultra OpenAI launches Dots, new ‘always-on agents’ you can assign tasks to Check out 9to5Google on YouTube for more news: Ben is a Senior Editor for 9to5Google. Find him on Twitter @NexusBen. Send tips to schoon@9to5g.com or encrypted to benschoon@protonmail.com. river

匹配字段正文

Agentic AI / Generative AI Bring coding agents, model customization, and CUDA-accelerated development to your everyday PC NVIDIA RTX Spark combines the GB10 system-on-a-chip with up to 128GB of unified memory to let developers run local coding agents, test, and deploy on a single PC. The platform delivers 4.6x faster coding-agent performance, 2.7x faster fine-tuning, and 1.8x faster code compilation compared with Intel Panther Lake 12Xe. NVIDIA partnered with Microsoft to expand the Windows on A

匹配字段正文

AIPOCH Open-Science released a new open-source desktop app that lets AI agents handle research tasks, from reading papers to analyzing data in Python and R. The company built the app to lower the barrier to research, letting anyone with curiosity and a research idea get involved. By saving the code, inputs, and steps that produced each artifact, it lets others rerun and analyze published papers. Researchers can use models from OpenAI, Anthropic, DeepSeek, and others, as well as local models. Pro

匹配字段正文

This is a big win for reproducible research. AIPOCH has released Open-Science, a free, open-source desktop app where AI agents run scientific analyses and keep the code behind every result. Their newest feature is the .science file, which packs a finished research session, including conversation branches, file versions, notebook records and verification evidence, into one file you can import on another computer. The recipient can trace how you reached each result, but the import opens read-only,

匹配字段正文

Artificial Intelligence Introducing Reflection Beam: an open-weight model for agentic coding on AWS Reflection AI introduced Beam, an open-weight model for agentic coding and long-horizon tasks such as maintaining code, using tools, interpreting results, and completing multi-step workflows. The open-weight release gives developers and researchers control over the runtime, deployment environment, and customization process. In this post, we explain Reflection’s vision and show how you can deploy B

匹配字段正文

unced that RTX Spark laptops are available globally from ASUS, Dell, HP, Lenovo, Microsoft and MSI, with prices starting at $2,599 and software support at launch spanning local AI agents, creative applications and major games. The systems are sold through the manufacturers’ websites and major retailers, with more systems to follow, and pricing and configurations differ by region. The launch lineup comprises the ASUS ProArt P14 and P16, Dell XPS 16, HP OmniBook Ultra and OmniBook X 14, Lenovo Yog

匹配字段正文

。 快手CEO程一笑曾在电话会中表示，可灵营收的高增长源于B端企业客户的API调用和P端付费会员订阅的双轮驱动。 这种选择，是快手有限的资源支持与可灵对自身模型能力自信双重驱动的结果，也为可灵走向独立融资埋下伏笔。但当可灵走向独立，面向市场的评估，它自然需要讲出更好的故事。 Kling 4.0全量上线后，能力接入MCP/CLI。也就是说，它可以被塞进AI编剧Agent、剪辑流水线、独立影视工作室等场景，正在获得更广泛的应用场景。 从3.0到4.0，可灵升级时间节点 但这并不能表示，可灵会获得可持续的增长收入，千亿的估值或许也仍然有待商榷。 02 「离开」快手，180亿美元的可灵冲刺上市 ㅤ 2026年5月，快手计划分拆可灵AI的消息不胫而走。随后快手官方回应称，正在评估相关重组方案，但尚未签署任何最终协议。7月2日，快手正式对外宣布了可灵引入外部融资的消息。 根据快手在港交所发布的公告，可灵敲定30亿美元外部增资，创全球视频大模型公司单轮融资纪录，投后估值180亿美元。 快手公告：可灵AI落地近30亿美元融资 消息公布后，快手股价一度大涨7%，但很快吐回涨幅。7月6日，腾讯减持快手，持

匹配字段正文

soning, Tool Calls and LoRA Optimizations Fusion Optimizations BF16 MoE Kernel Optimization TPU Support Reinforcement learning in Miles Validated Run Performance Results OpenHands Agentic Coding Sweep Reflection’s Beam is a 501B-parameter mixture-of-experts model with 23B active parameters per token, built for software engineering and long-horizon agentic tasks. With its planned Apache 2.0 release, Beam is designed for developers to deploy and build on across the runtimes, frameworks, and coding

匹配字段正文