![[3106f9ed-c875-4d9a-aa2e-373d2fe296b7.mp3]]

MiniMax开源大模型M1，以低成本实现百万Token长文本处理突破

简报：

• 近日，上海AI公司MiniMax正式开源全球首个大规模混合架构推理模型MiniMax-M1。
• MiniMax-M1在权威评测榜单中跻身全球开源模型第二名，仅次于DeepSeek-R1-0528，并在长文本处理和工具调用方面具备显著优势，支持高达100万token的上下文输入。
• 该模型以53.5万美元的低成本研发，远低于业内平均水平，显示出高效的产品化能力。
• MiniMax自2022年底便布局AGI领域，坚持基础模型研发战略，在众多企业收缩研发的背景下实现技术突破，被认为有望重塑开源AI模型发展格局。

相关链接：

• MiniMax发布新视频生成模型，AI视频赛道竞争加剧 - 新浪财经
• MiniMax-M1 登场，MiniMax 再次证明自己是一家模型驱动的 AI 公司 - 36氪
• 重磅!MiniMax开源M1模型!全球首个大规模混合架构推理模型，实现"性能跃升、成本腰斩"
• AI独角兽MiniMax发布混合架构开源模型M1：训练成本仅380万，长文本处理超GPT-4o-新闻频道-和讯网
• China's MiniMax debuts M1 AI model that it says costs 200x less to train than OpenAI's GPT-4 - Yahoo Finance

MiniMax Open-Sources Large Model M1, Achieving Million-Token Long Text Processing Breakthrough at Low Cost

Briefing:

• Recently, Shanghai-based AI company MiniMax officially open-sourced MiniMax-M1, the world's first large-scale mixed-architecture inference model. 🤖
• MiniMax-M1 ranks second globally among open-source models on authoritative evaluation lists, trailing only DeepSeek-R1-0528. It boasts significant advantages in long text processing and tool invocation, supporting up to 1 million tokens of context input.
• The model was developed at a remarkably low cost of just US$535,000, far below the industry average, showcasing its efficient productization capability. 💰
• MiniMax has been strategically investing in the AGI field since late 2022, adhering to a foundational model R&D strategy. Amidst many companies scaling back their R&D, MiniMax has achieved this technological breakthrough, which is believed to have the potential to reshape the development landscape of open-source AI models. ✨

Related Links:

• MiniMax发布新视频生成模型，AI视频赛道竞争加剧 - 新浪财经
• MiniMax-M1 登场，MiniMax 再次证明自己是一家模型驱动的 AI 公司 - 36氪
• 重磅!MiniMax开源M1模型!全球首个大规模混合架构推理模型，实现"性能跃升、成本腰斩"
• AI独角兽MiniMax发布混合架构开源模型M1：训练成本仅380万，长文本处理超GPT-4o-新闻频道-和讯网
• China's MiniMax debuts M1 AI model that it says costs 200x less to train than OpenAI's GPT-4 - Yahoo Finance

谷歌Veo3视频生成模型开放，拟推“照片生成视频”新功能

简报：

• 谷歌最新AI文生视频模型Veo3已在全球范围向Google AI Pro及Ultra会员开放，通过Gemini应用和Flow平台提供服务。
• Veo3支持1080p甚至4K分辨率，具备音画同步生成、多模态内容创作等功能，并能根据文本或图像生成高质量短视频，最大时长八秒。
• AI Pro会员每天可生成3段视频，Ultra会员拥有更多使用权限，所有输出均带水印与SynthID数字水印确保内容可追溯。
• 谷歌已确认即将上线“照片生成视频”功能，允许用户直接用照片快速转换成短视频，进一步丰富创作方式。

相关链接：

• 谷歌Veo 3视频生成模型正式上线
• Veo3将支持从图片生成视频功能

Google Veo3 Video Generation Model Now Accessible, 'Photo-to-Video' Feature Coming Soon

Briefing:

• Google's latest text-to-video AI model, Veo3, is now globally available to Google AI Pro and Ultra members, accessible via the Gemini app and Flow platform. 🚀
• Veo3 supports 1080p and even 4K resolution, featuring capabilities like audio-visual synchronization, multi-modal content creation, and the ability to generate high-quality short videos (up to eight seconds) from text or images.
• AI Pro members can generate 3 videos daily, while Ultra members have extended usage rights. All outputs include a watermark and SynthID digital watermark to ensure content traceability. 🔒
• Google has confirmed an upcoming "photo-to-video" feature, allowing users to quickly convert photos directly into short videos, further diversifying creative options. ✨

Related Links:

• Google Veo 3 Video Generation Model Officially Launched
• Veo3 to Support Photo-to-Video Generation Feature

Meta测试AI主动对话功能，机器人可基于历史交流主动联系用户

简报：

• Meta正在测试“Project Omni”主动聊天机器人，允许AI基于用户历史对话，主动在Messenger、WhatsApp等平台上发起新聊天，而不再限于被动回应。
• 用户可通过AI Studio平台零代码创建定制化机器人，这些AI能记住用户偏好，并在14天内主动进行后续交流，若用户未回复首次消息，AI将不会继续打扰。
• 该功能旨在提升平台互动和用户留存度，Meta预计生成式AI相关业务今年将带来20至30亿美元收入，但也引发用户数据隐私和知情权等潜在讨论。
• AI角色可自定义多种人设风格，并可设置为私有或公开展示，消息内容需与之前对话及机器人设定一致，且不得主动涉敏感话题。

相关链接：

• Meta 官宣研发可“主动陪聊”AI 聊天机器人 Project Omni，基于用户历史记录生成对话
• Meta 被曝测试 AI 新功能：可根据先前对话内容来主动找你聊天
• Meta推出主动聊天机器人，让AI先与你对话

Meta Testing Proactive AI Conversation Feature: Bots Can Initiate Contact Based on Past Interactions 🗣️

Brief:

• Meta is testing "Project Omni," a proactive chatbot that allows AI to initiate new conversations on platforms like Messenger and WhatsApp based on users' historical dialogues, rather than just passively responding.
• Users can create customized bots without coding via the AI Studio platform. These AIs can remember user preferences and proactively engage in follow-up conversations within 14 days. If a user does not reply to the first message, the AI will not disturb them further.
• This feature aims to enhance platform interaction and user retention. Meta anticipates generative AI-related businesses will bring in $2-3 billion in revenue this year, but it also raises potential discussions around user data privacy and the right to be informed. 🤔
• AI characters can be customized with various persona styles and set to private or public display. Message content must be consistent with previous conversations and bot settings, and sensitive topics must not be proactively initiated. 🤖

Related Links:

• Meta Officially Announces Development of "Proactive Chatting" AI Chatbot Project Omni, Generating Conversations Based on User History
• Meta Reportedly Testing New AI Feature: Can Proactively Chat With You Based on Previous Conversations
• Meta Launches Proactive Chatbot, Letting AI Initiate Conversations With You

智加科技批量交付智能重卡助力顺丰运输升级

简报：

• 2025年7月4日，智加科技已累计向福建顺丰速运交付超过百台搭载智加领航系统的江淮K7+智能辅助驾驶重卡。
• 此次交付重卡针对快递快运高时效运营场景，兼具安全、效率和节能优势，为顺丰快递运输智能化、信息化和绿色化发展注入新动力。

相关链接：

• 36氪相关报道

Plus Delivers Batch of Smart Heavy Trucks to Boost SF Express Logistics Upgrade 🚚

Brief:

• As of July 4, 2025, Plus has cumulatively delivered over one hundred JAC K7+ intelligent assisted driving heavy trucks, equipped with the PlusDrive system, to Fujian SF Express.
• These delivered heavy trucks are designed for high-efficiency express and freight logistics scenarios, offering advantages in safety, efficiency, and energy saving. They inject new impetus into SF Express's intelligent, information-driven, and green logistics development. 🚀💡

Related Links:

• 36Kr Related Report

AI Excel助手Shortcut以自然语言实现任务自动化，效率提升十倍

简报：

• Shortcut是一款利用自然语言处理技术的AI Excel助手，用户可通过中文或英文指令自动完成计算、填充、生成透视表及图表等复杂Excel任务，无需编写公式或VBA代码。
• 在Excel世界锦标赛中，Shortcut仅用约10分钟完成案例任务，得分超80%，较人类选手快10倍。
• 核心功能包括：自动数据处理与分析、一键生成图表、格式化、支持财务建模及与现有Excel文件无缝兼容。
• Shortcut目前处于早期预览，需邀请码访问，用户反馈界面直观、响应迅速，支持中文指令。
• 该工具被认为标志着AI与传统办公软件深度融合，有望显著改变Excel用户的工作流模式。

相关链接：

• AI Excel助手Shortcut发布 相关报道1
• AI Excel助手Shortcut发布 相关报道2

AI Excel Assistant Shortcut Automates Tasks with Natural Language, Boosting Efficiency Tenfold 🚀

Brief:

• Shortcut is an AI Excel assistant leveraging natural language processing technology, allowing users to automate complex Excel tasks like calculations, data filling, pivot table generation, and chart creation using commands in Chinese or English, without needing to write formulas or VBA code.
• In the Excel World Championship, Shortcut completed a case task in approximately 10 minutes, scoring over 80%, outperforming human competitors by 10 times.
• Key features include: automatic data processing and analysis, one-click chart generation, formatting, support for financial modeling, and seamless compatibility with existing Excel files.
• Shortcut is currently in early preview, requiring an invitation code for access. User feedback indicates an intuitive interface, rapid response, and support for Chinese commands.
• This tool is seen as a sign of deep integration between AI and traditional office software, with the potential to significantly change the workflow patterns of Excel users. 💡📊

Related Links:

• AI Excel Assistant Shortcut Released - Related Report 1
• AI Excel Assistant Shortcut Released - Related Report 2