![[005c48f7-3689-40b4-ae74-45bd73daf069.mp3]]

飞利浦推出8号Pro AI耳机，集成腾讯混元大模型支持多语言互译

简报：

• 飞利浦于2025年在西安发布8号Pro AI耳机，该产品搭载腾讯混元AI语言大模型。
• 耳机支持17种语言互译、27种方言识别及4大专业领域术语库，配备55dB深度主动降噪技术。
• 该耳机为商务精英人群打造，标志着飞利浦在AI耳机领域的正式布局。

相关链接：

• 36氪新闻报道
• Aibase新闻介绍

Philips Launches No. 8 Pro AI Earbuds, Integrating Tencent Hunyuan Large Model for Multi-Language Translation Support

Brief:

• Philips unveiled the No. 8 Pro AI Earbuds in Xi'an in 2025, a product powered by Tencent Hunyuan AI Language Large Model. 🚀
• The earbuds support two-way translation for 17 languages, recognition for 27 dialects, and boast a terminology database for 4 major professional fields, equipped with 55dB deep active noise cancellation technology. 🌐
• Designed for business elites, these earbuds mark Philips' official entry into the AI earbud market. ✨

Related Links:

• 36Kr News Report
• Aibase News Introduction

OpenAI推出专为复杂任务设计的Deep Research API新模型o3与o4-mini

简报：

• 2025年6月26日，OpenAI发布了全新Deep Research API模型o3-deep-research和o4-mini-deep-research，专注于自动化研究和复杂任务处理，支持企业与开发者高效完成网页搜索、数据分析、代码执行等工作。
• o3模型面向高复杂推理与精准分析，如金融分析、科学研究等任务，价格为每1000次调用10-40美元；o4-mini则强调性价比和高效大规模查询，定价2-8美元。
• Deep Research API要求开发者提供明确输入，直接输出结构化带引用报告，适合市场分析、学术研究等场景；还支持异步任务处理和webhook自动通知提升效率。
• 新模型仅在ChatGPT生态内开放，旨在平衡现实世界说服力与安全风险，未来计划将该API与更多专业数据源整合。
• 发布恰逢业界AI研究工具竞争升级，OpenAI通过高性能与易用性巩固其企业级市场地位。

相关链接：

• AIBase新闻1
• AIBase新闻2
• AIBase新闻3
• AIBase新闻4
• IT之家相关报道

OpenAI Launches New Deep Research API Models o3 and o4-mini, Designed for Complex Tasks

Briefing:

• On June 26, 2025, OpenAI released its new Deep Research API models, o3-deep-research and o4-mini-deep-research. These models are dedicated to automating research and handling complex tasks, enabling enterprises and developers to efficiently perform web searches, data analysis, code execution, and more. 🚀
• The o3 model is designed for highly complex reasoning and precise analysis, suitable for tasks such as financial analysis and scientific research, priced at $10-40 per 1,000 calls. The o4-mini, conversely, emphasizes cost-effectiveness and efficient large-scale queries, priced at$2-8. 💰
• The Deep Research API requires developers to provide clear inputs and directly outputs structured, cited reports, making it ideal for market analysis, academic research, and similar scenarios. It also supports asynchronous task processing and automatic webhook notifications to enhance efficiency. 📜
• The new models are exclusively available within the ChatGPT ecosystem, aiming to balance real-world persuasive power with safety risks. Future plans include integrating this API with more specialized data sources.
• This release coincides with escalating competition in AI research tools within the industry. OpenAI aims to solidify its position in the enterprise market through high performance and ease of use.

Related Links:

• AIBase News 1
• AIBase News 2
• AIBase News 3
• AIBase News 4
• ITHome Related Report

谷歌发布Gemma3n多模态模型，移动设备实现云端级AI功能

简报：

• 谷歌于本周五开源全新端侧多模态大模型Gemma3n，使手机、平板、笔记本等本地设备具备此前只能在云端体验的多模态AI能力。
• Gemma3n模型结构创新，包括MatFormer架构、每层嵌入（PLE）技术和KV Cache共享，大幅提升内存效率和处理速度，最低2GB内存即可运行。
• 模型支持图像、音频、视频、文本输入，覆盖140种文本语言与35种多模态理解语言，E4B版在LMArena评测中刷新同级参数模型纪录。
• 谷歌已在 Hugging Face 平台开源模型和权重，推动端侧AI生态发展，为移动应用和智能硬件等场景带来更多可能。

相关链接：

• Gemma3n发布详情（aibase.com）
• https://huggingface.co/collections/google/gemma-3n-685065323f5984ef315c93f4
• 官方文档说明

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

Brief:

• Google open-sourced its new on-device multimodal large model, Gemma3n, this Friday. This enables local devices such as phones, tablets, and laptops to possess multimodal AI capabilities previously only available in the cloud. 📱
• Gemma3n features innovative model architecture, including MatFormer, Per-Layer Embedding (PLE) technology, and KV Cache sharing. These innovations significantly boost memory efficiency and processing speed, allowing the model to run on as little as 2GB of RAM. 🚀
• The model supports image, audio, video, and text inputs, covering 140 text languages and 35 multimodal understanding languages. The E4B version notably set a new record for models with comparable parameters in LMArena evaluations.
• Google has open-sourced the model and its weights on the Hugging Face platform, aiming to foster the on-device AI ecosystem and unlock more possibilities for mobile applications, smart hardware, and other scenarios. 💡

Related Links:

• Gemma3n Release Details (aibase.com)
• https://huggingface.co/collections/google/gemma-3n-685065323f5984ef315c93f4
• Official Documentation

深圳高校研发“南科盘古”人形机器人，实现多项自主智能技术

简报：

• 南方科技大学机器人研究院近日自主研制并发布了首款人形机器人“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。
• 该机器人由研究院团队独立完成设计和系统集成，具备高度拟人仿生机械臂，集成人工智能，实现多模式传感与多模态感知。
• “南科盘古”拥有双臂协同操作、导航、物体分割识别、智能拍照及类人社交行为等功能，能够模拟人类手臂大部分动作并进行智能交互。
• 此次突破展示了国内高校在高端人形机器人研发领域的技术实力和创新能力。

相关链接：

• 36氪报道
• IT之家新闻1
• IT之家新闻2
• IT之家新闻3

Shenzhen University Develops "SUSTech Pangu" Humanoid Robot, Achieving Multiple Autonomous Intelligent Technologies

Brief:

• The Institute of Robotics at Southern University of Science and Technology (SUSTech) recently independently developed and unveiled its first humanoid robot, "SUSTech Pangu," marking it as the first humanoid robot fully developed by a university in the Shenzhen region. 🤖
• The robot's design and system integration were independently completed by the Institute's team. It features highly anthropomorphic bionic robotic arms, integrates artificial intelligence, and achieves multi-mode sensing and multi-modal perception.
• "SUSTech Pangu" boasts functionalities such as dual-arm collaborative operation, navigation, object segmentation and recognition, smart photography, and human-like social behaviors. It can simulate most human arm movements and engage in intelligent interaction. 🦾
• This breakthrough demonstrates the technical strength and innovative capabilities of domestic universities in the field of high-end humanoid robot development. ✨

Related Links:

• 36Kr Report
• IT Home News 1
• IT Home News 2
• IT Home News 3

MacWhisper集成NVIDIA Parakeet模型，大幅加速音频转录速度

简报：

• macOS应用MacWhisper在最新版本中集成了NVIDIA Parakeet转录模型，可在本地实现极快音频转文字转换，8秒即可转录30分钟播客内容。
• Parakeet模型官方宣称，配备A100、H100、T4或V100 GPU的硬件上，1秒钟可转录60分钟音频，极大提升效率。
• 此次更新由开发者Jordi Bruin与Argmax团队合作完成，用户无需复杂部署，即可便捷体验Parakeet带来的速度提升。
• 实测显示，3小时播客通过新版MacWhisper转录仅需1分22秒，大幅领先旧有模型。

相关链接：

• IT之家报道
• 9to5Mac原文

MacWhisper Integrates NVIDIA Parakeet Model, Dramatically Accelerating Audio Transcription Speed ⚡️

Briefing:

• The macOS application MacWhisper has integrated the NVIDIA Parakeet transcription model in its latest version, enabling incredibly fast local audio-to-text conversion. It can transcribe 30 minutes of podcast content in just 8 seconds.
• The Parakeet model officially claims that on hardware equipped with A100, H100, T4, or V100 GPUs, it can transcribe 60 minutes of audio in 1 second, significantly boosting efficiency.
• This update was a collaboration between developer Jordi Bruin and the Argmax team, allowing users to easily experience the speed improvements brought by Parakeet without complex deployment.
• Practical tests show that a 3-hour podcast can be transcribed by the new MacWhisper in just 1 minute and 22 seconds, far outperforming older models. 🚀⏱️

Related Links:

• IT之家 Report
• 9to5Mac Original Article