Group 1 - Keda Xunfei launched the "Transformable Voice Replication" technology based on the Spark Voice large model, allowing users to replicate any voice with high fidelity using just one recording, which is expected to revolutionize fields such as digital humans, audiobooks, and content creation [2] - Yushu Technology unveiled a full-body remote operation platform, demonstrating its G1 robot performing household tasks like washing dishes, mopping, organizing clothes, and serving water under human remote control [2] - Keda Xunfei released the Spark X1.5 deep reasoning large model, achieving over 93% efficiency compared to international competitors in end-to-end performance, with capabilities in language understanding, text generation, knowledge Q&A, logical reasoning, mathematical skills, and coding, while supporting over 130 languages and reaching 95% performance of GPT-5 [2] - Google's AI model series Gemini made significant progress with the preview version "gemini-3-pro-preview-11-2025" discovered on the Vertex AI platform, indicating a release this month with support for a massive context window of up to 1 million tokens [2][3] Group 2 - The preview model identifier "11-2025" clearly points to a release window in November, with Vertex AI being a core tool for Google's cloud AI development platform, capable of handling up to 200,000 tokens at standard level and expanding to 1 million tokens at advanced level [3]

1. 谷歌计划11月面向部分开发者和企业用户开放Gemini 3 Pro预览版，正式发布可能在12月； 2. 该模型上下文窗口可达100万tokens，适合处理长文档和复杂数据流水线，尤其适合AI研究者和对上下文容量有极高要求的团队； 3. Gemini 3 Pro的发布还与谷歌即将推出的图像生成模型Nano Banana 2有关 ， 或 同步亮相。 生成式AI 一、Gemini 3 Pro预览版灰度测试，1M上下文窗口或成亮点 https://mp.weixin.qq.com/s/SSFQQDCvMFBEEwQJt1vE2A 二、苹果或每年砸 10 亿美元，将谷歌Gemini 模型用于Siri 1. 苹果正接近与谷歌达成协议每年支付约10亿美元，换取Gemini模型为新版Siri提供摘要器和任务规划等核心AI能力； 2. Gemini将运行在苹果私有云计算服务器上，用户数据不会接触谷歌系统，谷歌提供的模型拥有1.2万亿参数，远超苹果现有的1500亿参 数模型； 3. 苹果AI团队正面临严重人才流失，自7月至今已有数十名核心成员跳槽，基础模型团队负责人庞若鸣被Meta以2亿美元挖走。 https://mp.w ...