单词学习：![[6063560c-52ab-4263-940b-719d05f33aaf.mp3]]

新闻播报：![[9e9d7271-87b6-4bf6-a8e0-fc6ec04e9afa.mp3]]

OpenAI发布实时语音模型GPT-Realtime，新增图像输入与电话集成功能

简报：

• OpenAI正式推出其最先进的语音模型GPT-Realtime，该模型采用单一的端到端Speech-to-Speech架构，可直接处理和生成音频，显著降低延迟。
• 新增图像输入功能，模型能够处理并描述图像内容，为语音交互增加视觉上下文。
• 模型能捕捉笑声、停顿等非语言信号，支持对话中无缝切换语言，并根据场景调整语气。
• 通过支持远程MCP和SIP协议，开发者可将GPT-Realtime集成至电话系统或外部工具，实现实时通话。
• Realtime API价格下调20%，音频输入降至每百万token 32美元，输出降至64美元。

相关链接：

• OpenAI发布GPT-Realtime语音模型，支持图像输入及语音AI Agent设计
• OpenAI 推出 gpt-realtime 语音对话模型：支持情感感知、多语言无缝切换

OpenAI Launches Real-time Voice Model GPT-Realtime, Adds Image Input and Phone Integration 🚀

Brief:

• OpenAI officially launches its most advanced voice model, GPT-Realtime. This model utilizes a single, end-to-end Speech-to-Speech architecture, directly processing and generating audio to significantly reduce latency.
• New image input capability allows the model to process and describe image content, adding visual context to voice interactions. 🖼️
• The model can capture non-verbal cues such as laughter and pauses, supports seamless language switching during conversations, and adjusts its tone based on the scenario.
• By supporting remote MCP and SIP protocols, developers can integrate GPT-Realtime into telephone systems or external tools for real-time calls.
• Realtime API prices have been reduced by 20%, with audio input now costing $32 per million tokens and output$64.

Related Links: 🔗

• OpenAI Launches GPT-Realtime Voice Model, Supports Image Input & Voice AI Agent Design
• OpenAI Introduces GPT-Realtime Voice Dialogue Model: Supports Emotion Perception, Seamless Multi-language Switching

苹果Xcode新测试版发布，集成Claude与GPT-5双AI模型

简报：

• 苹果发布了Xcode 26 Beta 7版本，正式集成了Anthropic公司的Claude Sonnet 4 AI模型。
• 开发者可通过添加自己的Claude付费账号，在Xcode内直接使用AI生成代码、编写文档、定位并自动修复错误。
• 新增“内联交互演练场”（inline playgrounds）功能，允许开发者在编辑器内实时运行AI生成的代码示例。
• 该版本同时将内置的ChatGPT支持升级至GPT-5模型，并作为默认选项。
• 开发者可在ChatGPT与Claude两大AI编码辅助工具中自由选择，相关费用通过用户自有的AI账户配额计费。

相关链接：

• 苹果Xcode重磅集成Claude Sonnet4
  开发迎来AI革命时代
• 苹果发布 Xcode 26 Beta 7：新增 GPT-5 支持并集成 Claude

Apple Xcode New Beta Released, Integrating Dual AI Models: Claude and GPT-5 🤖

Brief:

• Apple has released Xcode 26 Beta 7, officially integrating Anthropic's Claude Sonnet 4 AI model. 🚀
• Developers can now use AI to generate code, write documentation, and identify and automatically fix errors directly within Xcode by adding their own paid Claude account.
• A new "inline interactive playgrounds" feature allows developers to run AI-generated code examples in real-time within the editor.
• This version also upgrades the built-in ChatGPT support to the GPT-5 model, making it the default option.
• Developers can freely choose between the two major AI coding assistants, ChatGPT and Claude, with associated costs billed through the user's own AI account quotas. 💰

Related Links:

• Apple Xcode's Major Integration of Claude Sonnet 4: iOS Development Enters the Era of AI Revolution
• Apple Releases Xcode 26 Beta 7: Adds GPT-5 Support and Integrates Claude

百度AI搜索助手提速，回答速度提升5倍

简报： 百度搜索AI助手已全量上线自研的极速模型，该技术使其首个词的返回时间缩短了39%，回答吐字速度加快了500%。

相关链接：

• 百度搜索AI助手已全量上线极速模型
• 百度升级 AI 云基础设施，发布百舸 5.0 计算平台、千帆 4.0 企业级 AI 开发平台

Baidu AI Search Assistant Accelerates, Answer Speed Boosted by 5x 🚀

Brief: Baidu Search AI Assistant has fully launched its self-developed "Extreme Speed Model." This technology reduces the time to return the first word by 39% and accelerates the answer generation speed by 500%. 💨✨

Related links:

• Baidu Search AI Assistant Fully Launches Extreme Speed Model
• Baidu Upgrades AI Cloud Infrastructure, Releases Baige 5.0 Computing Platform, Qianfan 4.0 Enterprise AI Development Platform

辉达Q2财报超预期股价反跌，分析师反问“不然要卖到外星？”

简报：

• 辉达（NVIDIA）公布2026财年第二季财报，营收与获利均优于市场预期，但盘后股价因“利多出尽”而重挫。
• 市场批评主要集中于两大隐忧：第三季财测指引仅比市场共识高出1%；以及库存暴增至149.6亿美元，引发对旧款芯片H100、H200需求下滑的担忧。
• 针对市场批评，有分析师认为是投资者期待过高、吹毛求疵，并反呛：“不然是要（辉达）卖到别的星球吗？”
• 尽管股价下跌，高盛等机构仍看好辉达前景，维持“买进”评等并调高目标价，认为其AI晶片垄断地位依旧稳固。

相关链接：

• 辉达财报／地表最強還不夠？財報優仍被酸　他逆風挺：不然賣到別的星球
• 辉达财报／股价已涨无可涨？陸行之點破盤後暴跌原因　庫存量成最大隱憂
• 獨家／史上最強抽獎！輝達官方FB留言　黃仁勳親筆簽名 RTX 5090免費送

NVIDIA Q2 Earnings Beat Expectations, Stock Falls Conversely; Analyst Quips, "Otherwise, Should They Sell to Aliens?" 📉

Briefing:

• NVIDIA announced its Q2 FY2026 earnings, with both revenue and profit exceeding market expectations. However, the post-market stock price plummeted due to the "all good news priced in" effect.
• Market criticism primarily centered on two major concerns: the Q3 revenue guidance was only 1% higher than market consensus, and inventory surged to $14.96 billion, raising worries about declining demand for older chips like H100 and H200.
• Addressing the market's criticism, some analysts believe investors' expectations are excessively high and nitpicky, retorting, "Otherwise, should NVIDIA sell to other planets?" 🤔🚀
• Despite the stock price decline, institutions like Goldman Sachs remain optimistic about NVIDIA's future, maintaining a "Buy" rating and raising their price target, believing its monopolistic position in AI chips remains solid.

Related Links:

• NVIDIA Earnings / Is "Strongest on Earth" Not Enough? Earnings Beat Still Criticized; Analyst Defends: "Otherwise, Sell to Other Planets"
• NVIDIA Earnings / Has the Stock Price Risen as Far as It Can? Lu Xingzhi Pinpoints Post-Market Plunge Reason; Inventory Becomes Biggest Concern
• Exclusive / Strongest Lottery Ever! Comment on NVIDIA's Official FB; Jensen Huang's Autographed RTX 5090 Given Away for Free

xAI发布全新编码模型Grok Code Fast 1，多平台限时免费

简报：

• xAI于2025年8月28日发布专为软件开发设计的快速经济型大语言模型Grok Code Fast 1。
• 该模型采用全新轻量化架构，专注于速度和成本效益，擅长处理TypeScript、Python、Java、Rust等多种编程语言任务。
• 发布后提供为期一周的免费试用，用户可在GitHub Copilot、Cursor、Cline等多个主流编程平台中启用。
• 免费试用期结束后，模型将通过xAI API提供服务，定价为每百万输入令牌0.20美元，每百万输出令牌1.50美元。

相关链接：

• xAI官方发布Grok Code Fast 1
• xAI推出Grok Code Fast 1智能编码模型，限时免费开放

xAI Launches New Coding Model Grok Code Fast 1, Limited-Time Free Access Across Multiple Platforms 🚀

Brief:

• On August 28, 2025, xAI released Grok Code Fast 1, a fast and economical large language model designed specifically for software development. 💡
• The model employs a new lightweight architecture focused on speed and cost-efficiency, excelling at tasks across various programming languages including TypeScript, Python, Java, and Rust.
• A one-week free trial is available upon release, allowing users to enable it within multiple mainstream programming platforms such as GitHub Copilot, Cursor, and Cline. 🎉
• After the free trial period, the model will be accessible via the xAI API, priced at $0.20 per million input tokens and$1.50 per million output tokens.

Related Links:

• xAI Officially Releases Grok Code Fast 1
• xAI Launches Grok Code Fast 1 Intelligent Coding Model, Open for Limited-Time Free Access

微软发布自研AI模型，强化Copilot音频与文本能力

简报：

• 微软人工智能部门推出其首批两款自研AI模型：语音模型MAI-Voice-1和通用模型MAI-1-preview。
• MAI-Voice-1语音模型能在单块GPU上于1秒内生成1分钟时长的音频，已应用于“Copilot Daily”等功能中。
• MAI-1-preview通用模型旨在为Copilot未来的文本场景提供支持，训练过程动用了约1.5万块英伟达H100 GPU。
• 用户可在Copilot Labs平台体验MAI-Voice-1，而MAI-1-preview已在AI基准测试平台LMArena上进行公开测试。

相关链接：

• 微软推出自研AI模型MAI-Voice-1及MAI-1-preview，面向音频与Copilot文本场景

Microsoft Unveils Self-Developed AI Models, Boosting Copilot's Audio and Text Capabilities ✨

Brief:

• Microsoft's AI division has launched its first two self-developed AI models: the speech model MAI-Voice-1 and the general-purpose model MAI-1-preview.
• The MAI-Voice-1 speech model can generate 1 minute of audio in just 1 second on a single GPU 🎙️, and has already been integrated into features like "Copilot Daily."
• The MAI-1-preview general-purpose model is designed to support Copilot's future text scenarios 🧠, and its training utilized approximately 15,000 NVIDIA H100 GPUs.
• Users can experience MAI-Voice-1 on the Copilot Labs platform, while MAI-1-preview is already undergoing public testing on the AI benchmark platform LMArena 🚀.

Related Link:

• Microsoft Introduces Self-Developed AI Models MAI-Voice-1 and MAI-1-preview for Audio and Copilot Text Scenarios