单词学习：![[655c1889-4d62-4d15-a8d2-a5d8fa9e2df7.mp3]]

新闻播报：![[1979f82b-a610-496f-b99a-a19f3c9863a5.mp3]]

特斯拉确认百亿AI芯片订单，加速自研AI5与AI6芯片进程

简报：马斯克证实特斯拉AI5芯片已接近流片，将由台积电与三星共同生产，同时下一代AI6芯片设计已启动，并已与三星签订了价值164亿美元的代工协议。

相关连接：

• 特斯拉 AI 芯片加速：AI5 即将流片，AI6 设计启动，三星台积电双线并进

Tesla Confirms Multi-Billion AI Chip Order, Accelerating In-House AI5 and AI6 Chip Development 🚀

Brief: Elon Musk confirmed that Tesla's AI5 chip is nearing tape-out 💡, with production jointly handled by TSMC and Samsung. Concurrently, design for the next-generation AI6 chip has commenced, and a dJyH6kDkfC5RlM16.answerText6.4 billion foundry agreement has already been signed with Samsung 💰.

Related Link:

• 特斯拉 AI 芯片加速：AI5 即将流片，AI6 设计启动，三星台积电双线并进

小米发布MiMo-Embodied模型，一套权重打通自动驾驶与家用机器人

简报：

• 小米于2025年11月21日发布并开源了业界首个打通自动驾驶与具身智能的跨域基座模型MiMo-Embodied。
• 该模型使用同一套参数权重，能同时支持自动驾驶（环境感知、状态预测、驾驶规划）与具身智能（可供性推理、任务规划、空间理解）的核心任务。
• 在29项核心基准测试中，该模型全面刷新记录，其中17项具身智能基准和12项自动驾驶基准取得SOTA（State-Of-The-Art）成绩。
• 小米计划将该模型率先用于自研扫地机器人、工厂AGV及SU7高阶智驾系统，并预计于2026年第一季度起通过OTA推送。

相关链接：

• 小米开源跨域具身大模型MiMo-Embodied：29项基准刷新SOTA，自驾+家用场景一套权重通吃

Xiaomi Unveils MiMo-Embodied Model: One Set of Weights Bridges Autonomous Driving and Home Robotics 🤖🚗

Briefing:

• On November 21, 2025, Xiaomi launched and open-sourced MiMo-Embodied, the industry's first cross-domain foundational model that integrates autonomous driving and embodied AI.
• Using a single set of parameter weights, the model simultaneously supports core tasks in autonomous driving (environmental perception, state prediction, driving planning) and embodied AI (affordance reasoning, task planning, spatial understanding).
• Across 29 core benchmark tests, the model comprehensively broke records, achieving State-Of-The-Art (SOTA) results in 17 embodied AI benchmarks and 12 autonomous driving benchmarks. ✨
• Xiaomi plans to first deploy this model in its self-developed robot vacuum cleaners, factory AGVs, and the SU7 advanced intelligent driving system, with over-the-air (OTA) updates expected to commence in Q1 2026. 🚀

Related Link:

• Xiaomi Open-Sources Cross-Domain Embodied Large Model MiMo-Embodied: 29 Benchmarks Set New SOTA, One Set of Weights for Both Autonomous Driving + Home Scenarios

VSORA Jotunn8 处理器成功流片，成欧洲首款 HBM AI 芯片

简报：VSORA 宣布其 Jotunn8 处理器成功流片，作为欧洲首款配备 HBM 内存的数据中心 AI 推理处理器，该芯片采用台积电 5nm 工艺，FP8 精度算力可达 3200TFLOPS。

相关链接：

• 欧洲首款 HBM 内存数据中心推理处理器 VSORA Jotunn8 流片，台积电 5nm 制程

VSORA Jotunn8 Processor Successfully Taped Out, Becomes Europe's First HBM AI Chip 🇪🇺🎉

Brief: VSORA announced the successful tape-out of its Jotunn8 processor. As Europe's first data center AI inference processor equipped with HBM memory, the chip utilizes TSMC's 5nm process and delivers an FP8 precision computing power of up to 3200 TFLOPS! 🚀

Related Links:

• Europe's First HBM Memory Data Center Inference Processor VSORA Jotunn8 Taped Out, TSMC 5nm Process

Anthropic新模型Claude Opus4.5疑遭泄露，或于今日发布

简报：

• AI基准测试平台Epoch AI页面短暂出现一个代号为“Claude Kayak”的新模型条目，发布时间标注为今日（2025年11月24日），但随后被删除。
• 社区普遍认为“Claude Kayak”即Anthropic即将推出的旗舰模型Claude Opus4.5，旨在对标OpenAI的GPT-5.1和Google的Gemini3.0Pro。
• 该模型预计将在复杂推理、多步代理任务及代码生成能力上实现大幅提升，被视为2025年末AI领域最重磅的潜在发布之一。
• 截至目前，Anthropic官方尚未对此消息做出回应。

相关链接：

• Claude Opus4.5 或今日重磅发布，对标 GPT-5.1 和 Gemini Pro

Anthropic's New Model Claude Opus 4.5 Allegedly Leaked, Possibly Releasing Today 🔥

Briefing:

• A new model entry, codenamed "Claude Kayak," briefly appeared on the AI benchmarking platform Epoch AI's page, with a release date marked as today (November 24, 2025), but was subsequently removed.
• The community widely believes "Claude Kayak" is Anthropic's upcoming flagship model, Claude Opus 4.5, aiming to compete with OpenAI's GPT-5.1 and Google's Gemini 3.0 Pro.
• This model is anticipated to achieve significant advancements in complex reasoning, multi-step agent tasks, and code generation capabilities, positioning it as one of the most significant potential AI releases of late 2025. 🚀
• As of now, Anthropic has not officially responded to this news.

Related Links:

• Claude Opus 4.5 Possibly Set for Major Release Today, Aiming to Compete with GPT-5.1 and Gemini Pro

清华定义AI“能力密度”新指标：模型体积每3.5个月可减半

简报：

• 清华大学联合面壁智能在《自然・机器智能》发文，提出“能力密度”（Capability Density）新指标，指出AI模型性能应“重密度、轻规模”。
• 研究统计发现，AI模型的能力密度约每3.5个月翻倍，意味着实现同等性能所需的模型参数量可指数级下降。
• 面壁智能已基于该理论推出0.5B至2B参数的“高密度”模型，其性能可达到7B至13B规模模型的水平。
• 该系列小型化模型已成功在手机语音助手、车载交互及智能家居等终端设备上规模化落地，推理延迟低于100ms，功耗下降45%。

相关链接：

• 清华提出“能力密度”评估体系：AI模型每3. 5 个月体积减半，面壁智能已落地手机/车载/家居

Tsinghua Defines New AI Metric "Capability Density": Model Size Can Halve Every 3.5 Months

Briefing: 📝

• Tsinghua University, in collaboration with ModelBest, published an article in Nature Machine Intelligence, proposing a new metric called "Capability Density," emphasizing that AI model performance should prioritize "density over scale."
• Research statistics reveal that the capability density of AI models roughly doubles every 3.5 months, meaning the number of model parameters required to achieve equivalent performance can decrease exponentially.
• Based on this theory, ModelBest has launched "high-density" models ranging from 0.5B to 2B parameters, whose performance can match that of 7B to 13B scale models.
• This series of miniaturized models has been successfully implemented on terminal devices such as mobile voice assistants, in-car interactions, and smart homes, with inference latency below 100ms and power consumption reduced by 45%. 🚀

Related Link:

• Tsinghua Proposes "Capability Density" Evaluation System: AI Model Size Halves Every 3.5 Months, ModelBest Already Deployed on Phones/Vehicles/Homes 💡

OpenAI 宣布 GPT-4o API 将于明年2月停用，建议开发者迁移至GPT-5.1

简报：

• OpenAI 已通知所有 API 客户，chatgpt-4o-latest 接口将于 2026 年 2 月 16 日正式关闭，此后该模型仅保留在消费者端 ChatGPT 产品中。
• 本次停服仅限于 API 调用通道，网页版、移动端及桌面版 ChatGPT 不受影响。
• 官方建议开发者迁移至 GPT-5.1 系列模型，新模型在性能上更优且输入成本更低。
• 停服原因是 GPT-4o 的 API 调用量持续下滑，已被视为“遗留系统”，维持双轨运营会增加成本。
• 开发者需在截止日期前完成模型切换与测试，逾期调用将返回 410 Gone 错误。

相关链接：

• OpenAI 确认：GPT-4o API 将于 2026 年 2 月 16 日正式停服，开发者仅剩三个月迁移窗口

OpenAI Announces GPT-4o API Retirement by Next February, Recommends Developers Migrate to GPT-5.1

Briefing:

• OpenAI has notified all API customers that the chatgpt-4o-latest interface will officially be shut down on February 16, 2026. After this date, the model will only remain available in consumer-facing ChatGPT products. 🗓️
• This deprecation is limited strictly to the API calling channel; web, mobile, and desktop versions of ChatGPT will not be affected.
• The official recommendation is for developers to migrate to the GPT-5.1 series models, which offer superior performance and lower input costs. ✨
• The reason for the shutdown is the continuous decline in GPT-4o API call volume, leading it to be considered a "legacy system." Maintaining dual-track operations would increase costs.
• Developers are required to complete the model switch and testing before the deadline; calls made after this period will return a 410 Gone error. ⚠️

Related Links:

• OpenAI Confirms: GPT-4o API to Officially Cease Service on February 16, 2026, Developers Have Only Three Months for Migration

斑马智行AI合作成果显著，已携手宝马、大众等逾10家车企

简报：

• 斑马智行在成立十周年之际宣布，其AI业务在不到一年内已与智己、宝马、红旗、大众、现代等超过10个汽车品牌达成量产或定点合作。
• 公司计划在接下来的一年内，实现元神AI Agent的规模化落地，并推动端侧全模态大模型实车方案Auto Omni从定点向量产。

相关链接：

• 斑马智行：在AI领域已经量产或定点合作10余个汽车品牌

Zebra Zhixing Achieves Remarkable AI Cooperation, Partnering with BMW, Volkswagen, and Over 10 Automakers 🚀

Brief:

• Celebrating its tenth anniversary, Zebra Zhixing announced that its AI business has secured mass production or designated partnership agreements with over 10 automotive brands, including IM Motors, BMW, Hongqi, Volkswagen, and Hyundai, in less than a year. 🚗
• The company plans to achieve the large-scale deployment of its Yuan Shen AI Agent within the next year and to transition its edge-side full-modal large model in-vehicle solution, Auto Omni, from designated to mass production. ✨

Related Links:

• Zebra Zhixing: Over 10 Automotive Brands in Mass Production or Designated Partnership in AI Field