![[4626ee0a-9521-43ae-9b48-812023522e65.mp3]]

飞利浦推出新品8号Pro AI耳机，集成腾讯混元大模型支持多语互译

简报：

• 飞利浦影音及配件在2025新品发布会上推出8号Pro AI耳机，该产品搭载腾讯混元AI语言大模型。
• 该耳机支持17种语言互译、27种方言识别和4大专业领域术语库，配备55dB深度主动降噪功能。
• 此次发布还包括超薄磁吸充电宝及家庭娱乐KTV·云·音箱，标志飞利浦布局中国AI耳机等新赛道。

相关链接：

• 36氪新闻
• IT之家报道

Philips Unveils New 8 Pro AI Headphones, Integrates Tencent Hunyuan Large Model for Multilingual Translation Support

Brief:

• Philips Sound & Accessories unveiled the 8 Pro AI Headphones at its 2025 New Product Launch Event, powered by Tencent Hunyuan AI Large Language Model. 🚀
• The headphones support 17 languages for mutual translation, 27 dialect recognitions, and feature 4 major professional domain glossaries, along with 55dB deep active noise cancellation. 🗣️
• This launch also included an ultra-thin magnetic power bank and a home entertainment KTV Cloud Speaker, signaling Philips' strategic entry into new market segments in China, such as AI headphones. 🎶

Related Links:

• 36Kr News
• IT Home Report

Qwen VLo多模态视觉生成模型发布，支持多语言和开放式指令

简报：

• 通义千问近日正式发布Qwen VLo多模态大模型，支持图像内容的理解与高质量生成，并带来全新视觉创作体验。
• Qwen VLo在继承Qwen-VL系列优势的基础上升级，具备渐进式图片生成和优化机制，提升视觉效果与生成过程可控性。
• 该模型支持用户通过自然语言（中英等多语种）输入开放式指令，实现画风变化、元素添加、背景调整等多种创意修改。
• Qwen VLo可实现多图理解、图像检测与标注、文本到图像生成（支持任意分辨率和比例）以及艺术风格迁移等多样化功能。
• 目前Qwen VLo处于预览阶段，研发团队正不断优化以提升性能和稳定性，用户可在Qwen Chat平台体验。

相关链接：

• 通义千问发布多模态统一理解与生成模型Qwen VLo

Qwen VLo Multimodal Visual Generation Model Launched, Supporting Multi-language and Open-ended Instructions

Briefing:

• Tongyi Qianwen recently officially launched the Qwen VLo multimodal large model, supporting the understanding and high-quality generation of image content, and bringing a brand new visual creation experience. ✨
• Qwen VLo upgrades based on the strengths of the Qwen-VL series, featuring a progressive image generation and optimization mechanism, enhancing visual effects and control over the generation process.
• The model supports users in inputting open-ended instructions via natural language (multi-language, including Chinese and English) to achieve various creative modifications such as style changes, element additions, and background adjustments.
• Qwen VLo can perform diverse functions including multi-image understanding, image detection and annotation, text-to-image generation (supporting arbitrary resolutions and aspect ratios), and artistic style transfer. 🎨
• Qwen VLo is currently in the preview stage, and the R&D team is continuously optimizing it to improve performance and stability. Users can experience it on the Qwen Chat platform. 🚀

Related Links:

• Tongyi Qianwen Releases Multimodal Unified Understanding and Generation Model Qwen VLo

HeyGen推出AI视频Agent自动化内容创作流程

简报：

• HeyGen近期发布了一款AI视频Agent，能够实现素材上传后一键生成包括故事规划、脚本编写和镜头选择在内的完整视频制作流程；
• 该工具支持广告、短视频、产品演示等广泛应用场景，显著降低了内容创作门槛，提升创作效率；
• 用户无需具备专业视频编辑技能，即可通过直观界面快速获得高质量可发布视频，有助于品牌与个人在数字营销竞争中提升表现。

相关链接：

• HeyGen推出AI视频Agent 席卷内容创作界
• AI自动化视频生产解决方案
• 视频内容生产革命：HeyGen AI Agent

HeyGen Launches AI Video Agent to Automate Content Creation Workflow

Brief:

• HeyGen recently released an AI Video Agent 🚀 that enables a one-click full video production process, including story planning, script writing, and shot selection, after material upload.
• This tool supports a wide range of application scenarios such as advertisements, short videos, and product demonstrations, significantly lowering the barrier to content creation and improving production efficiency.
• Users can quickly obtain high-quality, publishable videos through an intuitive interface without needing professional video editing skills ✨, helping brands and individuals enhance their performance in the digital marketing competition. 🎬

Related Links:

• HeyGen's AI Video Agent Sweeps the Content Creation Scene
• AI Automated Video Production Solution
• Revolutionizing Video Content Production: HeyGen AI Agent

OpenAI推出o3与o4-mini Deep Research API模型，强化自动化研究与信息合成能力

简报：

• 2025年6月26日，OpenAI正式发布两款全新Deep Research API模型：o3-deep-research-2025-06-26与o4-mini-deep-research-2025-06-26，着重提升自动化网页搜索、数据分析和代码执行等任务能力。
• o3模型定位于复杂推理和高精度分析领域，推理性能最高，适合金融、科研等高要求任务；o4-mini强调效率与成本效益，适合大规模查询和快速整合信息。
• 新API要求开发者直接通过API输入清晰的提示，实现结构化带引用的报告输出，支持异步处理和webhook通知等特性，优化任务监控和工作流效率。
• Deep Research API目前仅限于ChatGPT生态使用，OpenAI正评估其潜在现实世界说服风险，并未完全开放至其他场景。
• 此举标志着OpenAI在企业自动化研究AI市场发力，面对谷歌Gemini Deep Research、DeepSeek等开源方案的竞争。

相关链接：

• aibase.com新闻稿1
• aibase.com新闻稿2
• TechRadar相关报道
• MSN报道

OpenAI Launches o3 and o4-mini Deep Research API Models, Strengthening Automated Research and Information Synthesis Capabilities

Briefing:

• On June 26, 2025, OpenAI officially released two new Deep Research API models: o3-deep-research-2025-06-26 and o4-mini-deep-research-2025-06-26, focusing on enhancing capabilities for tasks such as automated web search, data analysis, and code execution. 🚀
• The o3 model is positioned for complex reasoning and high-precision analysis, offering the highest inference performance, suitable for demanding tasks in finance, scientific research, and more. The o4-mini emphasizes efficiency and cost-effectiveness, ideal for large-scale queries and rapid information integration. 📊
• The new APIs require developers to provide clear prompts directly via the API, enabling structured, cited report outputs. They support features like asynchronous processing and webhook notifications, optimizing task monitoring and workflow efficiency.
• The Deep Research API is currently limited to the ChatGPT ecosystem. OpenAI is evaluating its potential real-world persuasion risks and has not fully opened it up for other scenarios. 🤔
• This move marks OpenAI's push into the enterprise automated research AI market, facing competition from solutions like Google Gemini Deep Research and open-source alternatives such as DeepSeek.

Related Links:

• aibase.com Press Release 1
• aibase.com Press Release 2
• TechRadar Related Report
• MSN Report

谷歌发布Gemma3n多模态模型，端侧设备实现云端级AI能力

简报：

• 2025年6月，谷歌开源了全新端侧多模态大模型Gemma3n，使手机、平板等设备具备此前仅能在云端完成的多模态AI处理能力。
• Gemma3n分为E2B和E4B两个版本，内存占用优化至2GB和3GB，支持图像、音频、视频、文本的多模态输入，可运行140种文本语言和35种多模态语言。
• 新架构采用Matryoshka Transformer、每层嵌入（PLE）技术、KV Cache共享及先进的音频/视觉编码器，实现灵活扩展与高内存效率，大幅提升推理及多模态处理速度。
• E4B模型在LMArena多项任务评测中超过1300分，为同尺寸模型的首例。
• 谷歌已在Hugging Face等平台开源Gemma3n模型和权重，支持移动AI、多语言交互、智能硬件等多领域开发，推动端侧AI应用创新。

相关链接：

• 谷歌Gemma3n多模态模型发布 - AIBASE新闻1
• Gemma3n技术解读 - AIBASE新闻2
• Hugging Face官方模型主页
• 谷歌官方文档

Google Releases Gemma3n Multimodal Model, Bringing Cloud-Level AI Capabilities to Edge Devices

Briefing:

• In June 2025, Google open-sourced its new edge-side multimodal large model, Gemma3n, enabling devices like smartphones and tablets to perform multimodal AI processing previously only possible in the cloud. 🚀
• Gemma3n comes in two versions, E2B and E4B, with optimized memory footprints of 2GB and 3GB respectively. It supports multimodal inputs including images, audio, video, and text, and can operate in 140 text languages and 35 multimodal languages.
• The new architecture incorporates Matryoshka Transformer, Per-Layer Embedding (PLE) technology, KV Cache sharing, and advanced audio/visual encoders. These innovations enable flexible scaling and high memory efficiency, significantly boosting inference and multimodal processing speeds. ✨
• The E4B model scored over 1300 points in multiple LMArena task evaluations, a first for a model of its size.
• Google has open-sourced the Gemma3n model and its weights on platforms like Hugging Face, supporting development across various fields such as mobile AI, multilingual interaction, and smart hardware, thereby fostering innovation in edge AI applications. 💡

Related Links:

• 谷歌Gemma3n多模态模型发布 - AIBASE新闻1
• Gemma3n技术解读 - AIBASE新闻2
• Hugging Face官方模型主页
• 谷歌官方文档

南科盘古：深圳首个高校自主研发人形机器人发布

简报：

• 南方科技大学机器人研究院近日推出首款自主研制的人形机器人“南科盘古”，成为深圳地区首个完全由高校独立研发的人形机器人。
• “南科盘古”由机器人研究院团队独立完成样机设计、系统集成和多模态智能交互体系，搭载自主研发的仿人机械臂和灵巧手系统，可实现双臂协同操作、导航、物体分割识别、智能拍照和类人社交行为。
• 该机器人集成人工智能技术，包括多模式传感和多模态大模型，在智能交互、视觉识别、感知与导航及拟人礼仪等方面展现高度仿生性能。

相关链接：

• 南方科技大学机器人研究院发布首款人形机器人“南科盘古”
• 南科盘古诞生：深圳高校首个完全自主研发人形机器人亮相

SUSTech Pangu: Shenzhen's First University-Developed Humanoid Robot Unveiled

Briefing:

• The Southern University of Science and Technology (SUSTech) Robotics Institute recently launched its first independently developed humanoid robot, "SUSTech Pangu," marking it as the first humanoid robot in the Shenzhen area to be entirely developed by a university. 🤖
• "SUSTech Pangu" saw its prototype design, system integration, and multimodal intelligent interaction system completed independently by the Robotics Institute team. It is equipped with independently developed human-like robotic arms and a dexterous hand system, enabling dual-arm collaborative operation, navigation, object segmentation and recognition, smart photography, and human-like social behaviors. 🦾
• The robot integrates artificial intelligence technologies, including multi-modal sensing and large multimodal models, demonstrating highly biomimetic performance in intelligent interaction, visual recognition, perception and navigation, and anthropomorphic etiquette. ✨

Related Links:

• Southern University of Science and Technology Robotics Institute releases first humanoid robot "SUSTech Pangu"
• SUSTech Pangu Born: Shenzhen University's First Fully Independently Developed Humanoid Robot Debuts