AI Daily: Qwen PC Client Launches AI Voice Input; ByteDance Releases Full-modal Large Model Doubao-Seed-2.0-lite; Google Updates AI Search Features

Welcome to the "AI Daily" section! This is your guide to exploring the world of artificial intelligence every day. Every day, we present you with the latest content in the AI field, focusing on developers to help you understand technology trends and learn about innovative AI product applications.

New AI products Click to learn more:https://app.aibase.com/en

1. Qwen PC Edition Launches AI Voice Input, Use It by Speaking Directly in Various Applications

The Qwen PC edition has launched an AI voice input function, allowing users to use it directly in various desktop applications through shortcut keys. It has powerful semantic parsing capabilities, capable of processing spoken content and organizing it in a structured way. It also supports voice commands to complete various office tasks, greatly improving work efficiency.

【AiBase Highlights:】

🗣️ Qwen voice input supports removing filler words, correcting errors, and formatting spoken content, and can provide intelligent responses based on context.

📝 Users can directly use Qwen for creation, Q&A, translation, etc., through voice commands in various applications.

📧 Qwen can automatically generate email replies, adapting to scenarios such as DingTalk, WeChat, or emails, enhancing office efficiency.

2. ByteDance Releases Full-modal Large Model Doubao-Seed-2.0-lite, AI Can Listen, Watch, and Even Work Directly

Doubao-Seed-2.0-lite, a large model released by ByteDance's Volcano Engine, achieves native unified understanding of video, image, audio, and text, marking a significant advancement in the field of multimodal interaction. The model performs outstandingly in visual and logical reasoning capabilities, especially excelling in complex reasoning tests in advanced subjects like physics and medicine, outperforming the previous Pro version. In addition, it has achieved integrated GUI understanding and execution for the first time, enabling operations such as clicking, dragging, and inputting, demonstrating strong AI capabilities.

【AiBase Highlights:】

✅ Achieves native unified understanding of video, image, audio, and text

🧠 Performs better than the Pro version in complex reasoning tests in advanced subjects like physics and medicine

🖱️ First implementation of integrated GUI understanding and execution, able to perform operations such as clicking, dragging, and inputting

3. Tencent Launches OpenSearch-VL: An All-in-One Solution for Open-source Multimodal Deep Search Agent

Tencent's HuanYuan, in collaboration with multiple universities, has released OpenSearch-VL, an open-source multimodal deep search intelligent agent solution that enhances model capabilities through reinforcement learning technology. The article details its innovative data production process, powerful tool environment, and fault-aware algorithms, showing excellent experimental performance. It is planned to be fully open-sourced, promoting the development of multimodal intelligent agent research.

【AiBase Highlights:】

🧠 Innovative data production line, overcoming the "search shortcut"

🛠 Powerful toolbox: Not just search

🔄 "Fault-aware" algorithm: Let the model learn from failure

Details link: https://arxiv.org/pdf/2605.05185

4. Moonshot AI Applies for Registration of KimiClaw Trademark, Or There May Be Major Hardware Moves?

Moonshot AI recently submitted multiple trademark registration applications for "KimiClaw," covering core areas such as scientific instruments, website services, and communication services, indicating the gradual exposure of its ambition in the AI ecosystem. The company was founded in 2023 by Yang Zhilin, focusing on the general AI field, and has already received $2 billion in funding, with a valuation expected to exceed $20 billion.

【AiBase Highlights:】

🧠 Moonshot AI applied for the registration of the "KimiClaw" trademark, covering the fields of scientific instruments, website services, and communication services.

🚀 The company was established for a short time but has already received $2 billion in funding, with a valuation expected to exceed $20 billion.

🔍 Trademark layout may indicate the company's ambition to expand from software algorithms to hardware devices or physical interactive products.

5. Mininglamp Opens Cider+Mano-P, Turning Your Mac into a Private AI Workstation

Mininglamp has opened up two local AI projects, Cider and Mano-P, which respectively solve the issues of Mac-side inference acceleration and GUI intelligent agent operation, providing users with a complete local AI workstation, improving efficiency and ensuring privacy and security.

【AiBase Highlights:】

🧠 Cider optimizes M-series chip performance, improving LLM/VLM inference speed and efficiency.

🖱️ Mano-P realizes pure visual GUI operation, supporting automation of complex desktop tasks.

🔒 Combining these two projects, it builds a local private AI infrastructure, ensuring privacy and security.

6. OpenAI Collaborates with Hardware Giants to Launch MRC Protocol, Aiming to End GPU Idle Waste

OpenAI collaborates with AMD, Intel, Microsoft, and NVIDIA to launch a new open network protocol, MRC, aimed at solving the efficiency bottlenecks of large-scale AI clusters, improving data transmission stability and reducing GPU idle waste, promoting the development of computing clusters towards a more efficient and green direction.

【AiBase Highlights:】

🧠 MRC protocol aims to optimize the performance of large-scale AI training clusters and improve data transmission stability.

⚡ Reduces GPU idle waste through multi-path connection schemes, improving computing efficiency.

🌐 OpenAI jointly launches MRC protocol with multiple industry giants, pushing large-scale computing clusters into an efficient and green new stage.

7. Google Updates AI Search Function: Integrates Reddit and Social Media First-hand Views

Google has made a major upgrade to its generative AI search function by integrating first-hand information sources such as social media, forums, and news subscriptions, optimizing the efficiency of users obtaining trusted information. This update introduces the "View Preview" feature, directly linking real-time online conversations from Reddit and various online forums to user queries. At the same time, expert advice is embedded in AI responses, and creator names and community nicknames are added to enhance the reliability of sources. Additionally, related topic extension links are displayed next to AI search results, such as directing users precisely to specific exposure suggestions on photography forums when searching for aurora photography techniques. For complex queries, Google has added suggested topics at the end of AI summaries, including case studies and blog reports, encouraging users to shift from single searches to in-depth research. Google also launched a subscription link highlighting function for news publishers, ensuring users can prioritize accessing subscribed authoritative information flows in AI mode. These series of strategic moves indicate that Google is trying to retain search traffic flowing to vertical social media by redefining source priorities.

【AiBase Highlights:】

🧠 Introduces the "View Preview" function, directly linking real-time conversations from Reddit and online forums to user queries.

📊 Embeds expert advice in AI responses and adds creator names and community nicknames to enhance the reliability of sources.

🌐 Adds extension links and suggested topics, encouraging users to shift from single searches to in-depth research.

8. xAI Launches Grok Imagine Quality Mode API: Visual Generation Enters a New Era of Realism

xAI has officially launched the "Quality Mode" of the Grok Imagine API, achieving significant breakthroughs in image realism, text rendering accuracy, and creative control, marking the beginning of a new era of realistic image generation.

【AiBase Highlights:】

🖼️ Enhances the delicacy of images, capturing natural skin texture, pore details, and complex lighting changes.

✍️ Solves character errors or formatting chaos in image generation models when handling complex text.

🚀 Enhances video generation capabilities, allowing brands to efficiently produce social media assets, product demonstration videos, and various commercial advertisements.