![[7eedca90-8ce6-4a66-aeca-3bed07da5873.mp3]]

OpenAI即将推出多模态GPT-5，7月上线在即

简报：

• OpenAI新一代AI模型GPT-5已进入灰度测试阶段，预计将在今年7月正式上线。
• GPT-5将支持文字、语音、图像、代码和视频等多模态输入，具备深度推理、实时视频生成和大规模代码编写能力。
• OpenAI CEO Sam Altman称，GPT-5是AI技术的重要飞跃，将整合推理与记忆以减少AI幻觉现象，极大拓展AI应用场景。
• 新一代模型旨在提供更自然直观的人机交互体验，为开发者和用户带来效率提升，推动AI成为生活和工作的操作系统。

相关链接：

• OpenAI多模态GPT-5即将上线
• ITHome报道GPT-5即将发布

OpenAI's Multimodal GPT-5 Launching Soon, Expected in July 🤖

Briefing:

• OpenAI's next-generation AI model, GPT-5, has entered the grey testing phase and is expected to officially launch in July this year.
• GPT-5 will support multimodal inputs including text, voice, images, code, and video, boasting capabilities in deep reasoning, real-time video generation, and large-scale code writing.
• OpenAI CEO Sam Altman stated that GPT-5 represents a significant leap in AI technology. It will integrate reasoning and memory to reduce AI hallucinations and vastly expand AI application scenarios. 🚀
• The new model aims to provide a more natural and intuitive human-AI interaction experience, bringing efficiency improvements for developers and users, and driving AI to become the operating system for life and work. 💡

Related Links:

• OpenAI's Multimodal GPT-5 Launching Soon
• ITHome Reports GPT-5 Imminent Release

百度文心大模型4.5系列全面开源，推动国内大模型领域创新升级

简报：

• 2025年6月30日，百度正式开源文心大模型4.5系列，共发布10款模型，包括47B、3B激活参数的混合专家（MoE）模型和0.3B参数的稠密模型，完整公开预训练权重及推理代码。
• 文心大模型4.5为原生多模态基础模型，不仅能理解文本，还可处理图片、视频等视觉信息，在多模态理解与生成上具备领先能力。
• 相关模型可于飞桨星河社区、Hugging Face等平台下载，并通过百度智能云千帆大模型平台提供API接入服务。
• 此次开源举措响应了国内大模型开源潮流，增强了行业技术交流和开发自由度，为开发者和研究者提供了更广泛的资源选择。
• 虽未包含4.5Turbo升级版，此次发布仍引发业界和开发者热议，有望与DeepSeek、阿里Qwen等主流模型展开竞争。

相关链接：

• 百度重磅开源文心大模型4.5系列
• 文心大模型4.5系列开源详情

Baidu ERNIE Bot 4.5 Series Fully Open-Sourced, Driving Innovation and Upgrade in Domestic Large Model Sector

Brief:

• On June 30, 2025, Baidu officially open-sourced the ERNIE Bot 4.5 series, releasing a total of 10 models. This includes Mixture-of-Experts (MoE) models with 47B and 3B active parameters, as well as a 0.3B dense model, with full disclosure of pre-training weights and inference code. 🚀
• The ERNIE Bot 4.5 is a native multi-modal foundational model, capable of not only understanding text but also processing visual information like images and videos, showcasing leading capabilities in multi-modal comprehension and generation.
• The relevant models are available for download on platforms such as PaddlePaddle Star River Community and Hugging Face, with API access services provided through Baidu AI Cloud Qianfan Large Model Platform.
• This open-sourcing initiative responds to the growing trend of open-source large models domestically, enhancing industry technical exchange and development freedom, and offering developers and researchers a wider range of resource options. 💡
• Although the 4.5 Turbo upgrade is not included, this release has still sparked widespread discussion among industry professionals and developers, and is expected to compete with mainstream models like DeepSeek and Alibaba's Qwen. 🔥

Related Links:

• 百度重磅开源文心大模型4.5系列
• 文心大模型4.5系列开源详情

OmniGen2开源多模态AI系统发布，文本与图像生成全面升级

简报：

• 北京人工智能研究院近日推出全新开源系统OmniGen2，实现文本到图像生成、图像编辑及上下文创作等多模态功能。
• OmniGen2采用独立的文本和图像解码路径，基于Qwen2.5-VL-3B多模态大模型，结合自定义扩散变换器，可支持多种艺术风格和多样化图文交互。
• 系统具备自我反思机制，能自动评估和改进生成图像，并在OmniContext基准测试中获得开源模型最高分，总分7.18。
• OmniGen2在文本提示、艺术风格适应及图像编辑等方面表现优异，但在图片清晰度和多图指令理解等仍有提升空间。
• 官方计划开放全部模型、训练数据和工具链，已在Hugging Face平台预发布开源资源。

相关链接：

• OmniGen2发布详情及开源入口
• 新闻参考1
• 新闻参考2

OmniGen2 Open-Source Multimodal AI System Released, Text and Image Generation Fully Upgraded

Briefing:

• The Beijing Institute for General Artificial Intelligence recently launched the new open-source system OmniGen2, achieving multimodal functionalities such as text-to-image generation, image editing, and context-aware creation. 🚀
• OmniGen2 utilizes independent text and image decoding pathways, built upon the Qwen2.5-VL-3B multimodal large model, combined with a custom diffusion transformer, enabling support for various artistic styles and diverse image-text interactions.
• The system features a self-reflection mechanism, allowing it to automatically evaluate and improve generated images. It achieved the highest score among open-source models in the OmniContext benchmark, with a total score of 7.18. 🌟
• OmniGen2 demonstrates excellent performance in text prompting, artistic style adaptation, and image editing; however, there is still room for improvement in image clarity and multi-image instruction comprehension.
• The official team plans to release all models, training data, and toolchains, with open-source resources already pre-released on the Hugging Face platform. 💻

Related Links:

• OmniGen2 Release Details and Open-Source Access
• News Reference 1
• News Reference 2

飞利浦发布8号Pro AI耳机，支持多语言实时互译和深度降噪

简报：

• 在2025年新品发布会上，飞利浦推出了8号Pro AI耳机，主打多语言沟通和高音质体验。
• 该耳机搭载腾讯混元AI语言大模型，可精准支持17种语言互译及27种方言识别，包含四大专业领域术语库，适用于商务会议及跨国交流场景。
• 配备55dB深度主动降噪技术，有效隔绝外部噪音，提升商务人士与年轻白领在嘈杂环境下的使用体验。
• 此次发布还包括“刀片”磁吸充电宝与KTV云音箱，彰显飞利浦在智能便携消费电子领域的创新布局。

相关链接：

• 飞利浦8号Pro AI耳机发布，支持多语言与主动降噪
• 飞利浦推8号Pro AI耳机，支持27种方言识别

Philips Unveils No. 8 Pro AI Headphones, Featuring Real-Time Multi-Language Translation and Deep Noise Cancellation

Briefing:

• At its 2025 new product launch event, Philips unveiled its No. 8 Pro AI Headphones, highlighting multi-language communication and a high-fidelity audio experience. 🚀
• Powered by Tencent's Hunyuan AI large language model, the headphones accurately support real-time translation for 17 languages and recognition for 27 dialects, including glossaries for four professional domains, making them ideal for business meetings and international communication scenarios.
• Equipped with 55dB deep active noise cancellation (ANC) technology, it effectively isolates external noise, enhancing the user experience for business professionals and young white-collar workers in noisy environments. 🎧
• The launch also included the "Blade" magnetic power bank and a KTV cloud speaker, showcasing Philips' innovative strategy in the smart portable consumer electronics sector. ✨

Related Links:

• Philips No. 8 Pro AI Headphones Launched, Supporting Multi-Language & Active Noise Cancellation
• Philips Launches No. 8 Pro AI Headphones, Supporting 27 Dialect Recognition

知乎直答上线知识库共享订阅功能推动专业内容开放

简报：

• 2025年6月30日，知乎直答发布新版知识库功能，支持内容共享订阅，并与社区深度融合，实现沉浸式阅读、精准提问、多文档提问等新体验
• 升级后的知识库将推动专业内容开放共享，提升AI搜索工具的专业度和可信度
• 公共知识库功能已在知乎直答网页端及App端全量上线

相关链接：

• 36氪报道
• AIBASE新闻

Zhihu Zhida Launches Knowledge Base Shared Subscription Feature, Promoting Open Access to Professional Content

Brief:

• On June 30, 2025, Zhihu Zhida released a new version of its knowledge base feature, supporting content sharing and subscription, and deeply integrating with the community to offer new experiences such as immersive reading, precise questioning, and multi-document queries. 🚀
• The upgraded knowledge base will promote the open sharing of professional content, enhancing the professionalism and credibility of AI search tools. 💡
• The public knowledge base feature has been fully launched on both the web and app versions of Zhihu Zhida. ✨

Related Links:

• 36Kr Report
• AIBASE News