AI Daily: Kimi K2.5 Launches; Alibaba Releases Inference Model Qwen3-Max-Thinking; Claude Deeply Integrates with Office Tools Like Slack

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 technical trends and innovative AI product applications.

Hot AI products Click to learn more:https://app.aibase.com/zh

1、Kimi K2.5 quietly launched with dual upgrades in vision and tool calling

The release of Kimi K2.5 marks Moonshot AI's continued efforts in the AI field, with significant improvements in visual and tool calling functions, greatly enhancing the model's practicality.

【AiBase Summary:】

🧠 Kimi K2.5 adds visual capabilities, supporting image analysis and 3D model generation.

🛠️ Tool calling function improved, making Kimi closer to "agent-style intelligence."

🚀 User feedback on performance is impressive, with high expectations for open-source.

2、Tencent Sogou Input Method releases 20.0 major version fully AI-powered

Tencent Sogou Input Method released version 20.0, a full AI upgrade, including three core features: AI voice, AI translation, and AI typing, significantly improving input efficiency and intelligent experience.

【AiBase Summary:】

🧠 AI voice input reduces delay by 40% and achieves an accuracy rate of 98%

🌐 AI translation supports 30 languages, achieving instant translation upon input.

💡 AI typing accuracy improved, with daily updates of popular vocabulary.

3、Domestic inference model sets a new benchmark! Alibaba releases Qwen3-Max-Thinking with trillions of parameters, performance approaching GPT-5.2

Alibaba released Qwen3-Max-Thinking, which has made a breakthrough in core dimensions such as complex reasoning, factual knowledge, and agent capabilities, with parameter count exceeding trillions and outstanding performance in multiple authoritative benchmark tests, approaching international top models.

【AiBase Summary:】

🧠 Trillion-level parameter inference: improves scientific knowledge and mathematical reasoning through larger-scale reinforcement learning and testing-time expansion techniques.

🤖 Native agent capabilities: the model has adaptive tool calling abilities and can autonomously choose search engines or code interpreters based on task requirements.

🚀 Performance comparable to international top models: performs well in 19 authoritative tests, matching GPT-5.2 and Claude-4.5, at the international leading level.

4、Ali Health's medical AI application "Hydrogen Ion" launches new features, supporting real-time tracking of global medical literature

Ali Health's medical AI application "Hydrogen Ion" released a major update, adding the "Dynamic Evidence Localization" feature. This feature upgrades traditional "static citations" to "live evidence," solving the pain point of AI misinformation (hallucination) in clinical and research scenarios. It accurately locates original sentences and synchronizes verification of the timeliness, authority, and logical consistency of information.

【AiBase Summary:】

✨ Added "Dynamic Evidence Localization" feature, enhancing the credibility and practicality of medical AI.

🔍 Supports accurate location of original sentences and synchronizes verification of timeliness, authority, and logical consistency of information.

🌐 Achieves real-time tracking of global medical guidelines and literature, facilitating doctors' access to complex clinical data.

5、Baidu Wenyi APP starts internal testing of the industry's first "multi-user multi-Agent" group chat

Baidu Wenyi APP started internal testing of the industry's first "multi-user multi-Agent" group chat, marking the entry of domestic AI applications into a new era of "multi-user + multi-AI" collaboration.

【AiBase Summary:】

🧠 Multiple agents coexist: multiple specialized agents can be called simultaneously in the group chat, forming a multidimensional "AI think tank."

🗣️ Proactive intervention mechanism: AI assistants can judge the timing based on the discussion atmosphere and actively provide suggestions.

🤝 Upgraded interaction paradigm: AI transforms from a passive response tool into a participant with social attributes.

6、Qwen3-Max-Thinking, the strongest domestic inference model, is now available on PC and web versions, stronger in initiative and skilled in logical reasoning

Alibaba released the strongest domestic inference model Qwen3-Max-Thinking, and Qwen AI assistant has integrated this model on PC and web versions, significantly improving reasoning capabilities and user experience.

【AiBase Summary:】

🧠 Qwen3-Max-Thinking is Alibaba's flagship inference model, with over trillion parameters and up to 36T Tokens of pre-training data.

💡 After integrating this model, Qwen AI assistant has stronger factual memory, complex reasoning, and self-validation capabilities, enhancing the user experience.

🌐 Qwen's monthly active users have exceeded 100 million and are fully integrated into Alibaba's ecosystem business, becoming the world's first AI assistant capable of completing real-life complex tasks.

7、Say goodbye to the chat window! Anthropic releases an interactive application, deeply integrating Slack and other office tools

Anthropic launched the Claude interactive application function, deeply integrating office tools to improve collaboration efficiency, while achieving ecological compatibility through the MCP protocol and emphasizing user security precautions.

【AiBase Summary:】

🧠 Claude integrates office tools such as Slack and Canva, improving work efficiency.

🔒 MCP protocol supports third-party service integration, enhancing AI ecological compatibility.

🛠️ Users need to be cautious with authorization to avoid leakage of sensitive information and ensure safe use.

8、Ant Group Lingbo opens a spatial perception model LingBot-Depth, enabling robots to accurately grasp transparent and reflective objects

Ant Group Lingbo Technology opened a high-precision spatial perception model LingBot-Depth, which has made breakthroughs in depth perception of transparent and reflective objects and performed well in authoritative evaluations. Its core data assets will help the community solve complex spatial perception problems in various scenarios.

【AiBase Summary:】

🧠 Using mask depth modeling technology, it enhances depth perception capabilities for transparent and reflective objects.

📊 In multiple benchmark evaluations, LingBot-Depth shows significantly better performance than existing models.

📦 Open-sourced 2M real-world depth data and 1M simulation data, promoting the community to overcome complex spatial perception challenges in various scenarios.