模型发布/更新
-
Grok Imagine 1.5 预览版发布
Grok Imagine 1.5 预览版已发布,即日起可在 API 中体验。SpaceXAI 正在发力。
-
MiniMax M3 1M token 解码加速 15.6 倍
15.6× faster decoding at 1M tokens 🔥 感谢 @FireworksAI_HQ 为 M3 提供推理支持。 立即尝试 👇
-
Ideogram v4.0 released with 2K resolution and JSON prompt support
Introducing Ideogram v4.0. Native 2K resolution, excellent text rendering, and JSON prompt support. Try it now in Krea.
产品发布/更新
-
Meta 面向 WhatsApp Business 的 AI 智能体现已全球上线
Meta 为 WhatsApp Business 打造的 AI 智能体面向全球商家开放,将按照模型 token 使用量向企业收费。
-
NousResearch releases public beta of Hermes Agent desktop app
The official Hermes Agent desktop app is now available!
-
xAI Grok语音模型上线Vapi平台
试试 @Vapi_AI 上最自然的TTS和性价比最高的STT API。 来自 @xai 的Grok STT和Grok TTS现已在企业语音AI平台Vapi上线。基于Vapi构建自定义语音智能体,可让它们用客户的语言交流、在受监管的工作流中…
-
Grok models land on Cloudflare AI Gateway
Try Grok models on @Cloudflare's AI Gateway!
-
OpenShell v0.0.55 发布:新增 Vertex AI 推理支持
OpenShell v0.0.55 🧩 Google Vertex AI 推理提供者 🔒 基于配置文件的策略可见性 🐳 网关中更好的 Podman 检测 🖥️ 恢复 GPU procfs 基准行为 🔧 CI 与文档修复 运行智能体对接 V…
-
OpenClaw 2026.6.1发布:新增Windows节点与技能工坊
OpenClaw 2026.6.1 已上线 🦞 🪟 原生 Windows 节点主机 🛠️ 用于自主学习型智能体的技能工坊(Skill Workshop) 📋 工作板(Workboard)编排 🧠 支持 MiniMax M3 Windows…
-
黄仁勋与纳德拉共议智能体AI时代
智能体AI时代来了。 从台北,黄仁勋与@satyanadella共同出席#MSBuild,展示NVIDIA与@Microsoft如何携手构建,从Windows设备到规模化AI工厂。 ▶️ 观看对话:https://nvda.ws/4uef…
行业动态
-
微软与OpenAI分道扬镳——如今双方准备正面交锋
微软与OpenAI的合作关系已彻底破裂,双方进入正面竞争态势。前DeepMind高管、现任微软AI主管Mustafa Suleyman明确表示,微软现在必须从头证明自己能独立完成所有必需的任务。这标志着两家科技巨头从紧密合作伙伴转变为直接…
-
Satya Nadella on his Microsoft Build keynote
Building the frontier intelligence ecosystem with us. Highlights from my keynote at Microsoft Build this morning.
-
欧盟公布全面技术主权计划,推动芯片与AI自主发展
欧盟公布一项全面技术主权计划,旨在扩大本土半导体、人工智能和云计算供应链,以减少对美国和亚洲的依赖。该计划覆盖芯片制造、AI基础设施及云计算服务等多个关键领域,目标是增强欧洲在核心技术上的自主能力。
-
Sensor Tower: OpenAI's ChatGPT passes 1 billion MAU, fastest ever
Market intelligence firm Sensor Tower estimates that OpenAI's ChatGPT surpassed 1 billion global monthly active users…
-
DeepSeek said to seek RMB 50 billion in first funding round, with Tencent, CATL among investors
DeepSeek plans to raise about RMB 50 billion in its first funding round from investors including Tencent and CATL.…
-
Suno closes $400 million Series D
We're excited to announce a new chapter for Suno: a $400 million Series D at a $5.4 billion valuation! 🚀 Our mission…
论文研究
-
NVIDIA Research 在 CVPR 2026 发表三篇论文:规模化训练实现抓取、自动驾驶与智能体泛化
NVIDIA Research 在 CVPR 2026 上发表三篇论文,展示规模化训练带来的泛化能力。GraspGen-X 是首个零样本抓取基础模型,基于 20 亿次模拟抓取训练,可为任意末端执行器生成抓取姿态。LCDrive 用紧凑潜在…
-
Anthropic analyzes 832 malicious AI accounts: medium- and high-risk attackers jump from 33% to 56% in six months
Anthropic analyzed 832 banned malicious accounts between March 2025 and March 2026, mapping them to the MITRE ATT&CK…
-
微软研究:装瓶厂AI从聊天到决策
一份在中西部装瓶厂进行的三个月试点显示,当AI超越聊天进入决策领域时会发生什么——约束条件变化、风险真实、答案必须可靠。 https://msft.it/6015vjYUN
技巧与观点
-
A functional taxonomy of world models
The World Labs team and Fei-Fei Li published an article sorting out the overused term "world model." Whereas language…
-
Anthropic powers self-serve data analytics with Claude
Anthropic used Claude to automate 95% of business analytics queries, with overall accuracy of about 95%. The key was…
-
A new chapter in flood resilience: Google open-sources hydrological modeling framework
Google Research has open-sourced its hydrological modeling framework on GitHub, a PyTorch-based Python package that…
-
Karpathy's llm-wiki project tops 5,000 stars
@karpathy's llm-wiki gained 5,000+ stars in a few weeks. The idea: stop rediscovering knowledge in every session. Let…
-
Reachy Mini 添加 MCP 工具
Reachy Mini 推出了公开的 MCP canary Space,用于远程工具调用。
-
DharmaOCR uses DPO to cut text degeneration rate by 59.4%
DharmaOCR, a structured OCR model released in April, used Direct Preference Optimization (DPO) as a second training…